Upload 122 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+melo/text/fr_phonemizer/example_ipa.txt filter=lfs diff=lfs merge=lfs -text

app.py

@@ -1,166 +1,182 @@
-import gradio as gr
-import os
-import warnings
-import asyncio
-from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Document, Settings
-from llama_index.llms.cerebras import Cerebras
-from llama_index.embeddings.mixedbreadai import MixedbreadAIEmbedding, EncodingFormat
-from groq import Groq
-import io
-
-# Suppress warnings
-warnings.filterwarnings("ignore", message=".*clean_up_tokenization_spaces.*")
-
-# Global variables
-index = None
-query_engine = None
-
-# Load Cerebras API key from Hugging Face secrets
-api_key = os.getenv("CEREBRAS_API_KEY")
-if not api_key:
-    raise ValueError("CEREBRAS_API_KEY is not set in Hugging Face Secrets.")
-else:
-    print("Cerebras API key loaded successfully.")
-
-# Initialize Cerebras LLM and embedding model
-os.environ["CEREBRAS_API_KEY"] = api_key
-llm = Cerebras(model="llama-3.3-70b", api_key=os.environ["CEREBRAS_API_KEY"])  # Change model to Llama3.1-70b from Cerebras
-Settings.llm = llm  # Ensure Cerebras is the LLM being used
-
-# Initialize Mixedbread Embedding model
-mixedbread_api_key = os.getenv("MXBAI_API_KEY")
-embed_model = MixedbreadAIEmbedding(api_key=mixedbread_api_key, model_name="mixedbread-ai/mxbai-embed-large-v1")
-
-# Initialize Groq client for Whisper Large V3
-groq_api_key = os.getenv("GROQ_API_KEY")
-if not groq_api_key:
-    raise ValueError("GROQ_API_KEY is not set.")
-else:
-    print("Groq API key loaded successfully.")
-client = Groq(api_key=groq_api_key)  # Groq client initialization
-
-# Function for audio transcription and translation (Whisper Large V3 from Groq)
-def transcribe_or_translate_audio(audio_file, translate=False):
-    """
-    Transcribes or translates audio using Whisper Large V3 via Groq API.
-    """
-    try:
-        with open(audio_file, "rb") as file:
-            if translate:
-                result = client.audio.translations.create(
-                    file=(audio_file, file.read()),
-                    model="whisper-large-v3",  # Use Groq Whisper Large V3
-                    response_format="json",
-                    temperature=0.0
-                )
-                return result.text
-            else:
-                result = client.audio.transcriptions.create(
-                    file=(audio_file, file.read()),
-                    model="whisper-large-v3",  # Use Groq Whisper Large V3
-                    response_format="json",
-                    temperature=0.0
-                )
-                return result.text
-    except Exception as e:
-        return f"Error processing audio: {str(e)}"
-
-# Function to load documents and create index
-def load_documents(file_objs):
-    global index, query_engine
-    try:
-        if not file_objs:
-            return "Error: No files selected."
-
-        documents = []
-        document_names = []
-        for file_obj in file_objs:
-            file_name = os.path.basename(file_obj.name)
-            document_names.append(file_name)
-            loaded_docs = SimpleDirectoryReader(input_files=[file_obj.name]).load_data()
-            for doc in loaded_docs:
-                doc.metadata["source"] = file_name
-                documents.append(doc)
-
-        if not documents:
-            return "No documents found in the selected files."
-
-        index = VectorStoreIndex.from_documents(documents, llm=llm, embed_model=embed_model)
-        query_engine = index.as_query_engine()
-
-        return f"Successfully loaded {len(documents)} documents from the files: {', '.join(document_names)}"
-    except Exception as e:
-        return f"Error loading documents: {str(e)}"
-
-async def perform_rag(query, history, audio_file=None, translate_audio=False):
-    global query_engine
-    if query_engine is None:
-        return history + [("Please load documents first.", None)]
-
-    try:
-        # Handle audio input if provided
-        if audio_file:
-            transcription = transcribe_or_translate_audio(audio_file, translate=translate_audio)
-            query = f"{query} {transcription}".strip()
-
-        response = await asyncio.to_thread(query_engine.query, query)
-        answer = str(response)  # Directly get the answer from the response
-
-        # If relevant documents are available, add sources without the "Sources" label
-        if hasattr(response, "get_documents"):
-            relevant_docs = response.get_documents()
-            if relevant_docs:
-                sources = "\n\n".join([f"{doc.metadata.get('source', 'No source available')}" for doc in relevant_docs])
-            else:
-                sources = ""
-        else:
-            sources = ""
-
-        # Combine answer with sources (if any) without additional labels
-        final_result = f"{answer}\n\n{sources}".strip()
-
-        # Return updated history with the final result
-        return history + [(query, final_result)]
-    except Exception as e:
-        return history + [(query, f"Error processing query: {str(e)}")]
-
-# Function to clear the session and reset variables
-def clear_all():
-    global index, query_engine
-    index = None
-    query_engine = None
-    return None, "", [], ""  # Reset file input, load output, chatbot, and message input to default states
-
-# Create the Gradio interface
-with gr.Blocks(theme=gr.themes.Base(primary_hue="teal", secondary_hue="teal", neutral_hue="slate")) as demo:
-    gr.Markdown("# RAG Multi-file Chat Application with Speech-to-Text")
-
-    chatbot = gr.Chatbot()
-
-    with gr.Row():
-        file_input = gr.File(label="Select files to load", file_count="multiple")
-        load_btn = gr.Button("Load Documents")
-        load_output = gr.Textbox(label="Load Status")
-
-    with gr.Row():
-        msg = gr.Textbox(label="Enter your question")
-        audio_input = gr.Audio(type="filepath", label="Upload Audio")
-        translate_checkbox = gr.Checkbox(label="Translate Audio to English Text", value=False)
-        
-    clear = gr.Button("Clear")
-
-    # Set up event handlers
-    load_btn.click(load_documents, inputs=[file_input], outputs=[load_output])
-
-    # Event handler for text input (only process text)
-    msg.submit(perform_rag, inputs=[msg, chatbot], outputs=[chatbot])
-
-    # Event handler for audio input (only process audio)
-    audio_input.change(perform_rag, inputs=[msg, chatbot, audio_input, translate_checkbox], outputs=[chatbot])
-
-    clear.click(clear_all, outputs=[file_input, load_output, chatbot, msg], queue=False)
-
-# Run the app
-if __name__ == "__main__":
-    demo.queue()
-    demo.launch()
+import gradio as gr
+import os
+import warnings
+import asyncio
+from melo.api import TTS
+from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Document, Settings
+from llama_index.llms.cerebras import Cerebras
+from llama_index.embeddings.mixedbreadai import MixedbreadAIEmbedding, EncodingFormat
+from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+from groq import Groq
+import io
+import nltk
+from dotenv import load_dotenv  # Import dotenv to load .env variables
+
+# Load environment variables from .env file
+load_dotenv()
+
+nltk.download('averaged_perceptron_tagger_eng')
+
+# Suppress warnings
+warnings.filterwarnings("ignore", message=".*clean_up_tokenization_spaces.*")
+
+# Global variables
+index = None
+query_engine = None
+
+# Inisialisasi MeloTTS untuk TTS
+device = 'cpu'  # Atur menjadi 'cuda' jika GPU tersedia
+language = 'EN'  # Bahasa default
+model = TTS(language=language, device=device)
+
+# Load Cerebras API key from environment
+api_key = os.getenv("CEREBRAS_API_KEY")
+if not api_key:
+    raise ValueError("CEREBRAS_API_KEY is not set in environment variables.")
+else:
+    print("Cerebras API key loaded successfully.")
+
+# Initialize Cerebras LLM and embedding model
+os.environ["CEREBRAS_API_KEY"] = api_key
+llm = Cerebras(model="llama-3.3-70b", api_key=os.environ["CEREBRAS_API_KEY"])  # Change model to Llama3.1-70b from Cerebras
+Settings.llm = llm  # Ensure Cerebras is the LLM being used
+
+# Initialize Mixedbread Embedding model
+mixedbread_api_key = os.getenv("MXBAI_API_KEY")
+embed_model = MixedbreadAIEmbedding(api_key=mixedbread_api_key, model_name="mixedbread-ai/mxbai-embed-large-v1")
+
+# Initialize Groq client for Whisper Large V3
+groq_api_key = os.getenv("GROQ_API_KEY")
+if not groq_api_key:
+    raise ValueError("GROQ_API_KEY is not set in environment variables.")
+else:
+    print("Groq API key loaded successfully.")
+client = Groq(api_key=groq_api_key)  # Groq client initialization
+
+# Function for audio transcription and translation (Whisper Large V3 from Groq)
+def transcribe_or_translate_audio(audio_file, translate=False):
+    """
+    Transcribes or translates audio using Whisper Large V3 via Groq API.
+    """
+    try:
+        with open(audio_file, "rb") as file:
+            if translate:
+                result = client.audio.translations.create(
+                    file=(audio_file, file.read()),
+                    model="whisper-large-v3",  # Use Groq Whisper Large V3
+                    response_format="json",
+                    temperature=0.0
+                )
+                return result.text
+            else:
+                result = client.audio.transcriptions.create(
+                    file=(audio_file, file.read()),
+                    model="whisper-large-v3",  # Use Groq Whisper Large V3
+                    response_format="json",
+                    temperature=0.0
+                )
+                return result.text
+    except Exception as e:
+        return f"Error processing audio: {str(e)}"
+
+# Function to load documents and create index
+def load_documents(file_objs):
+    global index, query_engine
+    try:
+        if not file_objs:
+            return "Error: No files selected."
+
+        documents = []
+        document_names = []
+        for file_obj in file_objs:
+            file_name = os.path.basename(file_obj.name)
+            document_names.append(file_name)
+            loaded_docs = SimpleDirectoryReader(input_files=[file_obj.name]).load_data()
+            for doc in loaded_docs:
+                doc.metadata["source"] = file_name
+                documents.append(doc)
+
+        if not documents:
+            return "No documents found in the selected files."
+
+        index = VectorStoreIndex.from_documents(documents, llm=llm, embed_model=embed_model)
+        query_engine = index.as_query_engine()
+
+        return f"Successfully loaded {len(documents)} documents from the files: {', '.join(document_names)}"
+    except Exception as e:
+        return f"Error loading documents: {str(e)}"
+
+async def perform_rag(query, history, audio_file=None, translate_audio=False):
+    global query_engine
+    if query_engine is None:
+        return history + [("Please load documents first.", None)], None  # Tambahkan None untuk output audio
+
+    try:
+        # Handle audio input jika diberikan
+        if audio_file:
+            transcription = transcribe_or_translate_audio(audio_file, translate=translate_audio)
+            query = f"{query} {transcription}".strip()
+
+        response = await asyncio.to_thread(query_engine.query, query)
+        answer = str(response)  # Dapatkan jawaban dari respons
+
+        # Jika dokumen relevan tersedia, tambahkan sumber tanpa label "Sources"
+        if hasattr(response, "get_documents"):
+            relevant_docs = response.get_documents()
+            if relevant_docs:
+                sources = "\n\n".join([f"{doc.metadata.get('source', 'No source available')}" for doc in relevant_docs])
+            else:
+                sources = ""
+        else:
+            sources = ""
+
+        # Gabungkan jawaban dengan sumber (jika ada) tanpa label tambahan
+        final_result = f"{answer}\n\n{sources}".strip()
+
+        # **Generate audio menggunakan MeloTTS**
+        output_audio_path = "output.wav"
+        model.tts_to_file(answer, model.hps.data.spk2id['EN-US'], output_audio_path, speed=1.0)
+
+        # Kembalikan history yang diperbarui dan file audio
+        return history + [(query, final_result)], output_audio_path
+    except Exception as e:
+        return history + [(query, f"Error processing query: {str(e)}")], None
+
+# Function to clear the session and reset variables
+def clear_all():
+    global index, query_engine
+    index = None
+    query_engine = None
+    return None, "", [], ""  # Reset file input, load output, chatbot, and message input to default states
+
+# Create the Gradio interface
+with gr.Blocks(theme=gr.themes.Base(primary_hue="teal", secondary_hue="teal", neutral_hue="slate")) as demo:
+    gr.Markdown("# RAG Multi-file Chat Application with Speech-to-Text and Text-to-Speech")
+    
+    chatbot = gr.Chatbot()
+    audio_output = gr.Audio(label="Response Audio", type="filepath")
+    with gr.Row():
+        file_input = gr.File(label="Select files to load", file_count="multiple")
+        load_btn = gr.Button("Load Documents")
+        load_output = gr.Textbox(label="Load Status")
+    with gr.Row():
+        msg = gr.Textbox(label="Enter your question")
+        audio_input = gr.Audio(type="filepath", label="RECORD")
+        translate_checkbox = gr.Checkbox(label="Translate Audio to English Text", value=False)
+    clear = gr.Button("Clear")
+
+    # Set up event handlers
+    load_btn.click(load_documents, inputs=[file_input], outputs=[load_output])
+
+    # Event handler untuk input teks (proses teks)
+    msg.submit(perform_rag, inputs=[msg, chatbot], outputs=[chatbot, audio_output])  # Tambahkan audio_output
+
+    # Event handler untuk input audio (proses audio)
+    audio_input.change(perform_rag, inputs=[msg, chatbot, audio_input, translate_checkbox], outputs=[chatbot, audio_output])  # Tambahkan audio_output
+
+    clear.click(clear_all, outputs=[file_input, load_output, chatbot, msg], queue=False)
+
+# Run the app
+if __name__ == "__main__":
+    demo.queue()
+    demo.launch()

melo/api.py

@@ -0,0 +1,135 @@
+import os
+import re
+import json
+import torch
+import librosa
+import soundfile
+import torchaudio
+import numpy as np
+import torch.nn as nn
+from tqdm import tqdm
+import torch
+
+from . import utils
+from . import commons
+from .models import SynthesizerTrn
+from .split_utils import split_sentence
+from .mel_processing import spectrogram_torch, spectrogram_torch_conv
+from .download_utils import load_or_download_config, load_or_download_model
+
+class TTS(nn.Module):
+    def __init__(self, 
+                language,
+                device='auto',
+                use_hf=True,
+                config_path=None,
+                ckpt_path=None):
+        super().__init__()
+        if device == 'auto':
+            device = 'cpu'
+            if torch.cuda.is_available(): device = 'cuda'
+            if torch.backends.mps.is_available(): device = 'mps'
+        if 'cuda' in device:
+            assert torch.cuda.is_available()
+
+        # config_path = 
+        hps = load_or_download_config(language, use_hf=use_hf, config_path=config_path)
+
+        num_languages = hps.num_languages
+        num_tones = hps.num_tones
+        symbols = hps.symbols
+
+        model = SynthesizerTrn(
+            len(symbols),
+            hps.data.filter_length // 2 + 1,
+            hps.train.segment_size // hps.data.hop_length,
+            n_speakers=hps.data.n_speakers,
+            num_tones=num_tones,
+            num_languages=num_languages,
+            **hps.model,
+        ).to(device)
+
+        model.eval()
+        self.model = model
+        self.symbol_to_id = {s: i for i, s in enumerate(symbols)}
+        self.hps = hps
+        self.device = device
+    
+        # load state_dict
+        checkpoint_dict = load_or_download_model(language, device, use_hf=use_hf, ckpt_path=ckpt_path)
+        self.model.load_state_dict(checkpoint_dict['model'], strict=True)
+        
+        language = language.split('_')[0]
+        self.language = 'ZH_MIX_EN' if language == 'ZH' else language # we support a ZH_MIX_EN model
+
+    @staticmethod
+    def audio_numpy_concat(segment_data_list, sr, speed=1.):
+        audio_segments = []
+        for segment_data in segment_data_list:
+            audio_segments += segment_data.reshape(-1).tolist()
+            audio_segments += [0] * int((sr * 0.05) / speed)
+        audio_segments = np.array(audio_segments).astype(np.float32)
+        return audio_segments
+
+    @staticmethod
+    def split_sentences_into_pieces(text, language, quiet=False):
+        texts = split_sentence(text, language_str=language)
+        if not quiet:
+            print(" > Text split to sentences.")
+            print('\n'.join(texts))
+            print(" > ===========================")
+        return texts
+
+    def tts_to_file(self, text, speaker_id, output_path=None, sdp_ratio=0.2, noise_scale=0.6, noise_scale_w=0.8, speed=1.0, pbar=None, format=None, position=None, quiet=False,):
+        language = self.language
+        texts = self.split_sentences_into_pieces(text, language, quiet)
+        audio_list = []
+        if pbar:
+            tx = pbar(texts)
+        else:
+            if position:
+                tx = tqdm(texts, position=position)
+            elif quiet:
+                tx = texts
+            else:
+                tx = tqdm(texts)
+        for t in tx:
+            if language in ['EN', 'ZH_MIX_EN']:
+                t = re.sub(r'([a-z])([A-Z])', r'\1 \2', t)
+            device = self.device
+            bert, ja_bert, phones, tones, lang_ids = utils.get_text_for_tts_infer(t, language, self.hps, device, self.symbol_to_id)
+            with torch.no_grad():
+                x_tst = phones.to(device).unsqueeze(0)
+                tones = tones.to(device).unsqueeze(0)
+                lang_ids = lang_ids.to(device).unsqueeze(0)
+                bert = bert.to(device).unsqueeze(0)
+                ja_bert = ja_bert.to(device).unsqueeze(0)
+                x_tst_lengths = torch.LongTensor([phones.size(0)]).to(device)
+                del phones
+                speakers = torch.LongTensor([speaker_id]).to(device)
+                audio = self.model.infer(
+                        x_tst,
+                        x_tst_lengths,
+                        speakers,
+                        tones,
+                        lang_ids,
+                        bert,
+                        ja_bert,
+                        sdp_ratio=sdp_ratio,
+                        noise_scale=noise_scale,
+                        noise_scale_w=noise_scale_w,
+                        length_scale=1. / speed,
+                    )[0][0, 0].data.cpu().float().numpy()
+                del x_tst, tones, lang_ids, bert, ja_bert, x_tst_lengths, speakers
+                # 
+            audio_list.append(audio)
+        torch.cuda.empty_cache()
+        audio = self.audio_numpy_concat(audio_list, sr=self.hps.data.sampling_rate, speed=speed)
+
+        if output_path is None:
+            return audio
+        else:
+            if format:
+                soundfile.write(output_path, audio, self.hps.data.sampling_rate, format=format)
+            else:
+                soundfile.write(output_path, audio, self.hps.data.sampling_rate)

melo/app.py

@@ -0,0 +1,61 @@
+# WebUI by mrfakename <X @realmrfakename / HF @mrfakename>
+# Demo also available on HF Spaces: https://huggingface.co/spaces/mrfakename/MeloTTS
+import gradio as gr
+import os, torch, io
+# os.system('python -m unidic download')
+print("Make sure you've downloaded unidic (python -m unidic download) for this WebUI to work.")
+from melo.api import TTS
+speed = 1.0
+import tempfile
+import click
+device = 'auto'
+models = {
+    'EN': TTS(language='EN', device=device),
+    'ES': TTS(language='ES', device=device),
+    'FR': TTS(language='FR', device=device),
+    'ZH': TTS(language='ZH', device=device),
+    'JP': TTS(language='JP', device=device),
+    'KR': TTS(language='KR', device=device),
+}
+speaker_ids = models['EN'].hps.data.spk2id
+
+default_text_dict = {
+    'EN': 'The field of text-to-speech has seen rapid development recently.',
+    'ES': 'El campo de la conversión de texto a voz ha experimentado un rápido desarrollo recientemente.',
+    'FR': 'Le domaine de la synthèse vocale a connu un développement rapide récemment',
+    'ZH': 'text-to-speech 领域近年来发展迅速',
+    'JP': 'テキスト読み上げの分野は最近急速な発展を遂げています',
+    'KR': '최근 텍스트 음성 변환 분야가 급속도로 발전하고 있습니다.',    
+}
+    
+def synthesize(speaker, text, speed, language, progress=gr.Progress()):
+    bio = io.BytesIO()
+    models[language].tts_to_file(text, models[language].hps.data.spk2id[speaker], bio, speed=speed, pbar=progress.tqdm, format='wav')
+    return bio.getvalue()
+def load_speakers(language, text):
+    if text in list(default_text_dict.values()):
+        newtext = default_text_dict[language]
+    else:
+        newtext = text
+    return gr.update(value=list(models[language].hps.data.spk2id.keys())[0], choices=list(models[language].hps.data.spk2id.keys())), newtext
+with gr.Blocks() as demo:
+    gr.Markdown('# MeloTTS WebUI\n\nA WebUI for MeloTTS.')
+    with gr.Group():
+        speaker = gr.Dropdown(speaker_ids.keys(), interactive=True, value='EN-US', label='Speaker')
+        language = gr.Radio(['EN', 'ES', 'FR', 'ZH', 'JP', 'KR'], label='Language', value='EN')
+        speed = gr.Slider(label='Speed', minimum=0.1, maximum=10.0, value=1.0, interactive=True, step=0.1)
+        text = gr.Textbox(label="Text to speak", value=default_text_dict['EN'])
+        language.input(load_speakers, inputs=[language, text], outputs=[speaker, text])
+    btn = gr.Button('Synthesize', variant='primary')
+    aud = gr.Audio(interactive=False)
+    btn.click(synthesize, inputs=[speaker, text, speed, language], outputs=[aud])
+    gr.Markdown('WebUI by [mrfakename](https://twitter.com/realmrfakename).')
+@click.command()
+@click.option('--share', '-s', is_flag=True, show_default=True, default=False, help="Expose a publicly-accessible shared Gradio link usable by anyone with the link. Only share the link with people you trust.")
+@click.option('--host', '-h', default=None)
+@click.option('--port', '-p', type=int, default=None)
+def main(share, host, port):
+    demo.queue(api_open=False).launch(show_api=False, share=share, server_name=host, server_port=port)
+
+if __name__ == "__main__":
+    main()

melo/attentions.py

@@ -0,0 +1,459 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from . import commons
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=4,
+        isflow=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+
+        self.cond_layer_idx = self.n_layers
+        if "gin_channels" in kwargs:
+            self.gin_channels = kwargs["gin_channels"]
+            if self.gin_channels != 0:
+                self.spk_emb_linear = nn.Linear(self.gin_channels, self.hidden_channels)
+                self.cond_layer_idx = (
+                    kwargs["cond_layer_idx"] if "cond_layer_idx" in kwargs else 2
+                )
+                assert (
+                    self.cond_layer_idx < self.n_layers
+                ), "cond_layer_idx should be less than n_layers"
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, g=None):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            if i == self.cond_layer_idx and g is not None:
+                g = self.spk_emb_linear(g.transpose(1, 2))
+                g = g.transpose(1, 2)
+                x = x + g
+                x = x * x_mask
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # pad along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

melo/commons.py

@@ -0,0 +1,160 @@
+import math
+import torch
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def intersperse(lst, item):
+    result = [item] * (len(lst) * 2 + 1)
+    result[1::2] = lst
+    return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def convert_pad_shape(pad_shape):
+    layer = pad_shape[::-1]
+    pad_shape = [item for sublist in layer for item in sublist]
+    return pad_shape
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

melo/configs/config.json

@@ -0,0 +1,94 @@
+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 52,
+    "epochs": 10000,
+    "learning_rate": 0.0003,
+    "betas": [
+      0.8,
+      0.99
+    ],
+    "eps": 1e-09,
+    "batch_size": 6,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 16384,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0,
+    "skip_optimizer": true
+  },
+  "data": {
+    "training_files": "",
+    "validation_files": "",
+    "max_wav_value": 32768.0,
+    "sampling_rate": 44100,
+    "filter_length": 2048,
+    "hop_length": 512,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 256,
+    "cleaned_text": true,
+    "spk2id": {}
+  },
+  "model": {
+    "use_spk_conditioned_encoder": true,
+    "use_noise_scaled_mas": true,
+    "use_mel_posterior_encoder": false,
+    "use_duration_discriminator": true,
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "n_layers_trans_flow": 3,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [
+      3,
+      7,
+      11
+    ],
+    "resblock_dilation_sizes": [
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ],
+      [
+        1,
+        3,
+        5
+      ]
+    ],
+    "upsample_rates": [
+      8,
+      8,
+      2,
+      2,
+      2
+    ],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [
+      16,
+      16,
+      8,
+      2,
+      2
+    ],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256
+  }
+}

melo/data/example/metadata.list

@@ -0,0 +1,20 @@
+data/example/wavs/000.wav|EN-default|EN|Well, there are always new trends and styles emerging in the fashion world, but I think some of the biggest trends at the moment include sustainability and ethical fashion, streetwear and athleisure, and oversized and deconstructed silhouettes.
+data/example/wavs/001.wav|EN-default|EN|Many designers and brands are focusing on creating more environmentally-friendly and socially responsible clothing, while others are incorporating elements of sportswear and casual wear into their collections.
+data/example/wavs/002.wav|EN-default|EN|And there's a growing interest in looser, more relaxed shapes and unconventional materials and finishes.
+data/example/wavs/003.wav|EN-default|EN|That's really insightful.
+data/example/wavs/004.wav|EN-default|EN|What do you think are some of the benefits of following fashion trends?
+data/example/wavs/005.wav|EN-default|EN|Well, I think one of the main benefits of following fashion trends is that it can be a way to express your creativity, personality, and individuality.
+data/example/wavs/006.wav|EN-default|EN|Fashion can be a powerful tool for self-expression and can help you feel more confident and comfortable in your own skin.
+data/example/wavs/007.wav|EN-default|EN|Additionally, staying up-to-date with fashion trends can help you develop your own sense of style and learn how to put together outfits that make you look and feel great.
+data/example/wavs/008.wav|EN-default|EN|That's a great point.
+data/example/wavs/009.wav|EN-default|EN|Do you think it's important to stay on top of the latest fashion trends, or is it more important to focus on timeless style?
+data/example/wavs/010.wav|EN-default|EN|I think it's really up to each individual to decide what approach to fashion works best for them.
+data/example/wavs/011.wav|EN-default|EN|Some people prefer to stick with classic, timeless styles that never go out of fashion, while others enjoy experimenting with new and innovative trends.
+data/example/wavs/012.wav|EN-default|EN|Ultimately, fashion is about personal expression and there's no right or wrong way to approach it.
+data/example/wavs/013.wav|EN-default|EN|The most important thing is to wear what makes you feel good and confident.
+data/example/wavs/014.wav|EN-default|EN|I completely agree.
+data/example/wavs/015.wav|EN-default|EN|Some popular ones that come to mind are oversized blazers, statement sleeves, printed maxi dresses, and chunky sneakers.
+data/example/wavs/016.wav|EN-default|EN|It's been really interesting chatting with you about fashion.
+data/example/wavs/017.wav|EN-default|EN|That's a good point.
+data/example/wavs/018.wav|EN-default|EN|What do you think are some current fashion trends that are popular right now?
+data/example/wavs/019.wav|EN-default|EN|There are so many trends happening right now, it's hard to keep track of them all!

melo/data_utils.py

@@ -0,0 +1,413 @@
+import os
+import random
+import torch
+import torch.utils.data
+from tqdm import tqdm
+from loguru import logger
+import commons
+from mel_processing import spectrogram_torch, mel_spectrogram_torch
+from utils import load_filepaths_and_text
+from utils import load_wav_to_torch_librosa as load_wav_to_torch
+from text import cleaned_text_to_sequence, get_bert
+import numpy as np
+
+"""Multi speaker version"""
+
+
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+    1) loads audio, speaker_id, text pairs
+    2) normalizes text and converts them to sequences of integers
+    3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths_sid_text, hparams):
+        self.audiopaths_sid_text = load_filepaths_and_text(audiopaths_sid_text)
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length = hparams.filter_length
+        self.hop_length = hparams.hop_length
+        self.win_length = hparams.win_length
+        self.sampling_rate = hparams.sampling_rate
+        self.spk_map = hparams.spk2id
+        self.hparams = hparams
+        self.disable_bert = getattr(hparams, "disable_bert", False)
+
+        self.use_mel_spec_posterior = getattr(
+            hparams, "use_mel_posterior_encoder", False
+        )
+        if self.use_mel_spec_posterior:
+            self.n_mel_channels = getattr(hparams, "n_mel_channels", 80)
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 300)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_sid_text)
+        self._filter()
+
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_sid_text_new = []
+        lengths = []
+        skipped = 0
+        logger.info("Init dataset...")
+        for item in tqdm(
+            self.audiopaths_sid_text
+        ):
+            try:
+                _id, spk, language, text, phones, tone, word2ph = item
+            except:
+                print(item)
+                raise
+            audiopath = f"{_id}"
+            if self.min_text_len <= len(phones) and len(phones) <= self.max_text_len:
+                phones = phones.split(" ")
+                tone = [int(i) for i in tone.split(" ")]
+                word2ph = [int(i) for i in word2ph.split(" ")]
+                audiopaths_sid_text_new.append(
+                    [audiopath, spk, language, text, phones, tone, word2ph]
+                )
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+            else:
+                skipped += 1
+        logger.info(f'min: {min(lengths)}; max: {max(lengths)}' )
+        logger.info(
+            "skipped: "
+            + str(skipped)
+            + ", total: "
+            + str(len(self.audiopaths_sid_text))
+        )
+        self.audiopaths_sid_text = audiopaths_sid_text_new
+        self.lengths = lengths
+
+    def get_audio_text_speaker_pair(self, audiopath_sid_text):
+        # separate filename, speaker_id and text
+        audiopath, sid, language, text, phones, tone, word2ph = audiopath_sid_text
+
+        bert, ja_bert, phones, tone, language = self.get_text(
+            text, word2ph, phones, tone, language, audiopath
+        )
+
+        spec, wav = self.get_audio(audiopath)
+        sid = int(getattr(self.spk_map, sid, '0'))
+        sid = torch.LongTensor([sid])
+        return (phones, spec, wav, sid, tone, language, bert, ja_bert)
+
+    def get_audio(self, filename):
+        audio_norm, sampling_rate = load_wav_to_torch(filename, self.sampling_rate)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError(
+                "{} {} SR doesn't match target {} SR".format(
+                    filename, sampling_rate, self.sampling_rate
+                )
+            )
+        # NOTE: normalize has been achieved by torchaudio
+        # audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if self.use_mel_spec_posterior:
+            spec_filename = spec_filename.replace(".spec.pt", ".mel.pt")
+        try:
+            spec = torch.load(spec_filename)
+            assert False
+        except:
+            if self.use_mel_spec_posterior:
+                spec = mel_spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.n_mel_channels,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    self.hparams.mel_fmin,
+                    self.hparams.mel_fmax,
+                    center=False,
+                )
+            else:
+                spec = spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    center=False,
+                )
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text, word2ph, phone, tone, language_str, wav_path):
+        phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str)
+        if self.add_blank:
+            phone = commons.intersperse(phone, 0)
+            tone = commons.intersperse(tone, 0)
+            language = commons.intersperse(language, 0)
+            for i in range(len(word2ph)):
+                word2ph[i] = word2ph[i] * 2
+            word2ph[0] += 1
+        bert_path = wav_path.replace(".wav", ".bert.pt")
+        try:
+            bert = torch.load(bert_path)
+            assert bert.shape[-1] == len(phone)
+        except Exception as e:
+            print(e, wav_path, bert_path, bert.shape, len(phone))
+            bert = get_bert(text, word2ph, language_str)
+            torch.save(bert, bert_path)
+            assert bert.shape[-1] == len(phone), phone
+
+        if self.disable_bert:
+            bert = torch.zeros(1024, len(phone))
+            ja_bert = torch.zeros(768, len(phone))
+        else:
+            if language_str in ["ZH"]:
+                bert = bert
+                ja_bert = torch.zeros(768, len(phone))
+            elif language_str in ["JP", "EN", "ZH_MIX_EN", "KR", 'SP', 'ES', 'FR', 'DE', 'RU']:
+                ja_bert = bert
+                bert = torch.zeros(1024, len(phone))
+            else:
+                raise
+                bert = torch.zeros(1024, len(phone))
+                ja_bert = torch.zeros(768, len(phone))
+        assert bert.shape[-1] == len(phone)
+        phone = torch.LongTensor(phone)
+        tone = torch.LongTensor(tone)
+        language = torch.LongTensor(language)
+        return bert, ja_bert, phone, tone, language
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def __getitem__(self, index):
+        return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_sid_text)
+
+
+class TextAudioSpeakerCollate:
+    """Zero-pads model inputs and targets"""
+
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text, audio and speaker identities
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized, sid]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]), dim=0, descending=True
+        )
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+        sid = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        tone_padded = torch.LongTensor(len(batch), max_text_len)
+        language_padded = torch.LongTensor(len(batch), max_text_len)
+        bert_padded = torch.FloatTensor(len(batch), 1024, max_text_len)
+        ja_bert_padded = torch.FloatTensor(len(batch), 768, max_text_len)
+
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        tone_padded.zero_()
+        language_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        bert_padded.zero_()
+        ja_bert_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, : text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, : spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, : wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+            sid[i] = row[3]
+
+            tone = row[4]
+            tone_padded[i, : tone.size(0)] = tone
+
+            language = row[5]
+            language_padded[i, : language.size(0)] = language
+
+            bert = row[6]
+            bert_padded[i, :, : bert.size(1)] = bert
+
+            ja_bert = row[7]
+            ja_bert_padded[i, :, : ja_bert.size(1)] = ja_bert
+
+        return (
+            text_padded,
+            text_lengths,
+            spec_padded,
+            spec_lengths,
+            wav_padded,
+            wav_lengths,
+            sid,
+            tone_padded,
+            language_padded,
+            bert_padded,
+            ja_bert_padded,
+        )
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
+    """
+
+    def __init__(
+        self,
+        dataset,
+        batch_size,
+        boundaries,
+        num_replicas=None,
+        rank=None,
+        shuffle=True,
+    ):
+        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+        print('buckets:', self.num_samples_per_bucket)
+
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+
+        try:
+            for i in range(len(buckets) - 1, 0, -1):
+                if len(buckets[i]) == 0:
+                    buckets.pop(i)
+                    self.boundaries.pop(i + 1)
+            assert all(len(bucket) > 0 for bucket in buckets)
+        # When one bucket is not traversed
+        except Exception as e:
+            print("Bucket warning ", e)
+            for i in range(len(buckets) - 1, -1, -1):
+                if len(buckets[i]) == 0:
+                    buckets.pop(i)
+                    self.boundaries.pop(i + 1)
+
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (
+                total_batch_size - (len_bucket % total_batch_size)
+            ) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+
+    def __iter__(self):
+        # deterministically shuffle based on epoch
+        g = torch.Generator()
+        g.manual_seed(self.epoch)
+
+        indices = []
+        if self.shuffle:
+            for bucket in self.buckets:
+                indices.append(torch.randperm(len(bucket), generator=g).tolist())
+        else:
+            for bucket in self.buckets:
+                indices.append(list(range(len(bucket))))
+
+        batches = []
+        for i in range(len(self.buckets)):
+            bucket = self.buckets[i]
+            len_bucket = len(bucket)
+            if len_bucket == 0:
+                continue
+            ids_bucket = indices[i]
+            num_samples_bucket = self.num_samples_per_bucket[i]
+
+            # add extra samples to make it evenly divisible
+            rem = num_samples_bucket - len_bucket
+            ids_bucket = (
+                ids_bucket
+                + ids_bucket * (rem // len_bucket)
+                + ids_bucket[: (rem % len_bucket)]
+            )
+
+            # subsample
+            ids_bucket = ids_bucket[self.rank :: self.num_replicas]
+
+            # batching
+            for j in range(len(ids_bucket) // self.batch_size):
+                batch = [
+                    bucket[idx]
+                    for idx in ids_bucket[
+                        j * self.batch_size : (j + 1) * self.batch_size
+                    ]
+                ]
+                batches.append(batch)
+
+        if self.shuffle:
+            batch_ids = torch.randperm(len(batches), generator=g).tolist()
+            batches = [batches[i] for i in batch_ids]
+        self.batches = batches
+
+        assert len(self.batches) * self.batch_size == self.num_samples
+        return iter(self.batches)
+
+    def _bisect(self, x, lo=0, hi=None):
+        if hi is None:
+            hi = len(self.boundaries) - 1
+
+        if hi > lo:
+            mid = (hi + lo) // 2
+            if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
+                return mid
+            elif x <= self.boundaries[mid]:
+                return self._bisect(x, lo, mid)
+            else:
+                return self._bisect(x, mid + 1, hi)
+        else:
+            return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size

melo/download_utils.py

@@ -0,0 +1,67 @@
+import torch
+import os
+from . import utils
+from cached_path import cached_path
+from huggingface_hub import hf_hub_download
+
+DOWNLOAD_CKPT_URLS = {
+    'EN': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/EN/checkpoint.pth',
+    'EN_V2': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/EN_V2/checkpoint.pth',
+    'FR': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/FR/checkpoint.pth',
+    'JP': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/JP/checkpoint.pth',
+    'ES': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/ES/checkpoint.pth',
+    'ZH': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/ZH/checkpoint.pth',
+    'KR': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/KR/checkpoint.pth',
+}
+
+DOWNLOAD_CONFIG_URLS = {
+    'EN': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/EN/config.json',
+    'EN_V2': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/EN_V2/config.json',
+    'FR': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/FR/config.json',
+    'JP': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/JP/config.json',
+    'ES': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/ES/config.json',
+    'ZH': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/ZH/config.json',
+    'KR': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/KR/config.json',
+}
+
+PRETRAINED_MODELS = {
+    'G.pth': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/pretrained/G.pth',
+    'D.pth': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/pretrained/D.pth',
+    'DUR.pth': 'https://myshell-public-repo-host.s3.amazonaws.com/openvoice/basespeakers/pretrained/DUR.pth',
+}
+
+LANG_TO_HF_REPO_ID = {
+    'EN': 'myshell-ai/MeloTTS-English',
+    'EN_V2': 'myshell-ai/MeloTTS-English-v2',
+    'EN_NEWEST': 'myshell-ai/MeloTTS-English-v3',
+    'FR': 'myshell-ai/MeloTTS-French',
+    'JP': 'myshell-ai/MeloTTS-Japanese',
+    'ES': 'myshell-ai/MeloTTS-Spanish',
+    'ZH': 'myshell-ai/MeloTTS-Chinese',
+    'KR': 'myshell-ai/MeloTTS-Korean',
+}
+
+def load_or_download_config(locale, use_hf=True, config_path=None):
+    if config_path is None:
+        language = locale.split('-')[0].upper()
+        if use_hf:
+            assert language in LANG_TO_HF_REPO_ID
+            config_path = hf_hub_download(repo_id=LANG_TO_HF_REPO_ID[language], filename="config.json")
+        else:
+            assert language in DOWNLOAD_CONFIG_URLS
+            config_path = cached_path(DOWNLOAD_CONFIG_URLS[language])
+    return utils.get_hparams_from_file(config_path)
+
+def load_or_download_model(locale, device, use_hf=True, ckpt_path=None):
+    if ckpt_path is None:
+        language = locale.split('-')[0].upper()
+        if use_hf:
+            assert language in LANG_TO_HF_REPO_ID
+            ckpt_path = hf_hub_download(repo_id=LANG_TO_HF_REPO_ID[language], filename="checkpoint.pth")
+        else:
+            assert language in DOWNLOAD_CKPT_URLS
+            ckpt_path = cached_path(DOWNLOAD_CKPT_URLS[language])
+    return torch.load(ckpt_path, map_location=device)
+
+def load_pretrain_model():
+    return [cached_path(url) for url in PRETRAINED_MODELS.values()]

melo/infer.py

@@ -0,0 +1,25 @@
+import os
+import click
+from melo.api import TTS
+
+    
+    
+@click.command()
+@click.option('--ckpt_path', '-m', type=str, default=None, help="Path to the checkpoint file")
+@click.option('--text', '-t', type=str, default=None, help="Text to speak")
+@click.option('--language', '-l', type=str, default="EN", help="Language of the model")
+@click.option('--output_dir', '-o', type=str, default="outputs", help="Path to the output")
+def main(ckpt_path, text, language, output_dir):
+    if ckpt_path is None:
+        raise ValueError("The model_path must be specified")
+    
+    config_path = os.path.join(os.path.dirname(ckpt_path), 'config.json')
+    model = TTS(language=language, config_path=config_path, ckpt_path=ckpt_path)
+    
+    for spk_name, spk_id in model.hps.data.spk2id.items():
+        save_path = f'{output_dir}/{spk_name}/output.wav'
+        os.makedirs(os.path.dirname(save_path), exist_ok=True)
+        model.tts_to_file(text, spk_id, save_path)
+
+if __name__ == "__main__":
+    main()

melo/init_downloads.py

@@ -0,0 +1,14 @@
+
+
+if __name__ == '__main__':
+
+    from melo.api import TTS
+    device = 'auto'
+    models = {
+        'EN': TTS(language='EN', device=device),
+        'ES': TTS(language='ES', device=device),
+        'FR': TTS(language='FR', device=device),
+        'ZH': TTS(language='ZH', device=device),
+        'JP': TTS(language='JP', device=device),
+        'KR': TTS(language='KR', device=device),
+    }

melo/losses.py

@@ -0,0 +1,58 @@
+import torch
+
+
+def feature_loss(fmap_r, fmap_g):
+    loss = 0
+    for dr, dg in zip(fmap_r, fmap_g):
+        for rl, gl in zip(dr, dg):
+            rl = rl.float().detach()
+            gl = gl.float()
+            loss += torch.mean(torch.abs(rl - gl))
+
+    return loss * 2
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+    loss = 0
+    r_losses = []
+    g_losses = []
+    for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+        dr = dr.float()
+        dg = dg.float()
+        r_loss = torch.mean((1 - dr) ** 2)
+        g_loss = torch.mean(dg**2)
+        loss += r_loss + g_loss
+        r_losses.append(r_loss.item())
+        g_losses.append(g_loss.item())
+
+    return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+    loss = 0
+    gen_losses = []
+    for dg in disc_outputs:
+        dg = dg.float()
+        l = torch.mean((1 - dg) ** 2)
+        gen_losses.append(l)
+        loss += l
+
+    return loss, gen_losses
+
+
+def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
+    """
+    z_p, logs_q: [b, h, t_t]
+    m_p, logs_p: [b, h, t_t]
+    """
+    z_p = z_p.float()
+    logs_q = logs_q.float()
+    m_p = m_p.float()
+    logs_p = logs_p.float()
+    z_mask = z_mask.float()
+
+    kl = logs_p - logs_q - 0.5
+    kl += 0.5 * ((z_p - m_p) ** 2) * torch.exp(-2.0 * logs_p)
+    kl = torch.sum(kl * z_mask)
+    l = kl / torch.sum(z_mask)
+    return l

melo/main.py

@@ -0,0 +1,36 @@
+import click
+import warnings
+import os
+
+
+@click.command
+@click.argument('text')
+@click.argument('output_path')
+@click.option("--file", '-f', is_flag=True, show_default=True, default=False, help="Text is a file")
+@click.option('--language', '-l', default='EN', help='Language, defaults to English', type=click.Choice(['EN', 'ES', 'FR', 'ZH', 'JP', 'KR'], case_sensitive=False))
+@click.option('--speaker', '-spk', default='EN-Default', help='Speaker ID, only for English, leave empty for default, ignored if not English. If English, defaults to "EN-Default"', type=click.Choice(['EN-Default', 'EN-US', 'EN-BR', 'EN_INDIA', 'EN-AU']))
+@click.option('--speed', '-s', default=1.0, help='Speed, defaults to 1.0', type=float)
+@click.option('--device', '-d', default='auto', help='Device, defaults to auto')
+def main(text, file, output_path, language, speaker, speed, device):
+    if file:
+        if not os.path.exists(text):
+            raise FileNotFoundError(f'Trying to load text from file due to --file/-f flag, but file not found. Remove the --file/-f flag to pass a string.')
+        else:
+            with open(text) as f:
+                text = f.read().strip()
+    if text == '':
+        raise ValueError('You entered empty text or the file you passed was empty.')
+    language = language.upper()
+    if language == '': language = 'EN'
+    if speaker == '': speaker = None
+    if (not language == 'EN') and speaker:
+        warnings.warn('You specified a speaker but the language is English.')
+    from melo.api import TTS
+    model = TTS(language=language, device=device)
+    speaker_ids = model.hps.data.spk2id
+    if language == 'EN':
+        if not speaker: speaker = 'EN-Default'
+        spkr = speaker_ids[speaker]
+    else:
+        spkr = speaker_ids[list(speaker_ids.keys())[0]]
+    model.tts_to_file(text, spkr, output_path, speed=speed)

melo/mel_processing.py

@@ -0,0 +1,174 @@
+import torch
+import torch.utils.data
+import librosa
+from librosa.filters import mel as librosa_mel_fn
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    output = dynamic_range_compression_torch(magnitudes)
+    return output
+
+
+def spectral_de_normalize_torch(magnitudes):
+    output = dynamic_range_decompression_torch(magnitudes)
+    return output
+
+
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    if torch.min(y) < -1.1:
+        print("min value is ", torch.min(y))
+    if torch.max(y) > 1.1:
+        print("max value is ", torch.max(y))
+
+    global hann_window
+    dtype_device = str(y.dtype) + "_" + str(y.device)
+    wnsize_dtype_device = str(win_size) + "_" + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device
+        )
+
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1),
+        (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+        mode="reflect",
+    )
+    y = y.squeeze(1)
+
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spectrogram_torch_conv(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    global hann_window
+    dtype_device = str(y.dtype) + '_' + str(y.device)
+    wnsize_dtype_device = str(win_size) + '_' + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
+
+    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    
+    # ******************** original ************************#
+    # y = y.squeeze(1)
+    # spec1 = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+    #                   center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+
+    # ******************** ConvSTFT ************************#
+    freq_cutoff = n_fft // 2 + 1
+    fourier_basis = torch.view_as_real(torch.fft.fft(torch.eye(n_fft)))
+    forward_basis = fourier_basis[:freq_cutoff].permute(2, 0, 1).reshape(-1, 1, fourier_basis.shape[1])
+    forward_basis = forward_basis * torch.as_tensor(librosa.util.pad_center(torch.hann_window(win_size), size=n_fft)).float()
+
+    import torch.nn.functional as F
+
+    # if center:
+    #     signal = F.pad(y[:, None, None, :], (n_fft // 2, n_fft // 2, 0, 0), mode = 'reflect').squeeze(1)
+    assert center is False
+
+    forward_transform_squared = F.conv1d(y, forward_basis.to(y.device), stride = hop_size)
+    spec2 = torch.stack([forward_transform_squared[:, :freq_cutoff, :], forward_transform_squared[:, freq_cutoff:, :]], dim = -1)
+
+
+    # ******************** Verification ************************#
+    spec1 = torch.stft(y.squeeze(1), n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
+                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+    assert torch.allclose(spec1, spec2, atol=1e-4)
+
+    spec = torch.sqrt(spec2.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    global mel_basis
+    dtype_device = str(spec.dtype) + "_" + str(spec.device)
+    fmax_dtype_device = str(fmax) + "_" + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=spec.dtype, device=spec.device
+        )
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+    return spec
+
+
+def mel_spectrogram_torch(
+    y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False
+):
+    global mel_basis, hann_window
+    dtype_device = str(y.dtype) + "_" + str(y.device)
+    fmax_dtype_device = str(fmax) + "_" + dtype_device
+    wnsize_dtype_device = str(win_size) + "_" + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=y.dtype, device=y.device
+        )
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device
+        )
+
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1),
+        (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+        mode="reflect",
+    )
+    y = y.squeeze(1)
+
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
+
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+
+    spec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    spec = spectral_normalize_torch(spec)
+
+    return spec

melo/models.py

@@ -0,0 +1,1030 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from melo import commons
+from melo import modules
+from melo import attentions
+
+from torch.nn import Conv1d, ConvTranspose1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+
+from melo.commons import init_weights, get_padding
+import melo.monotonic_align as monotonic_align
+
+
+class DurationDiscriminator(nn.Module):  # vits2
+    def __init__(
+        self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.gin_channels = gin_channels
+
+        self.drop = nn.Dropout(p_dropout)
+        self.conv_1 = nn.Conv1d(
+            in_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_1 = modules.LayerNorm(filter_channels)
+        self.conv_2 = nn.Conv1d(
+            filter_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_2 = modules.LayerNorm(filter_channels)
+        self.dur_proj = nn.Conv1d(1, filter_channels, 1)
+
+        self.pre_out_conv_1 = nn.Conv1d(
+            2 * filter_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.pre_out_norm_1 = modules.LayerNorm(filter_channels)
+        self.pre_out_conv_2 = nn.Conv1d(
+            filter_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.pre_out_norm_2 = modules.LayerNorm(filter_channels)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, in_channels, 1)
+
+        self.output_layer = nn.Sequential(nn.Linear(filter_channels, 1), nn.Sigmoid())
+
+    def forward_probability(self, x, x_mask, dur, g=None):
+        dur = self.dur_proj(dur)
+        x = torch.cat([x, dur], dim=1)
+        x = self.pre_out_conv_1(x * x_mask)
+        x = torch.relu(x)
+        x = self.pre_out_norm_1(x)
+        x = self.drop(x)
+        x = self.pre_out_conv_2(x * x_mask)
+        x = torch.relu(x)
+        x = self.pre_out_norm_2(x)
+        x = self.drop(x)
+        x = x * x_mask
+        x = x.transpose(1, 2)
+        output_prob = self.output_layer(x)
+        return output_prob
+
+    def forward(self, x, x_mask, dur_r, dur_hat, g=None):
+        x = torch.detach(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.conv_1(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_1(x)
+        x = self.drop(x)
+        x = self.conv_2(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_2(x)
+        x = self.drop(x)
+
+        output_probs = []
+        for dur in [dur_r, dur_hat]:
+            output_prob = self.forward_probability(x, x_mask, dur, g)
+            output_probs.append(output_prob)
+
+        return output_probs
+
+
+class TransformerCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        n_flows=4,
+        gin_channels=0,
+        share_parameter=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+
+        self.wn = (
+            attentions.FFT(
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+                isflow=True,
+                gin_channels=self.gin_channels,
+            )
+            if share_parameter
+            else None
+        )
+
+        for i in range(n_flows):
+            self.flows.append(
+                modules.TransformerCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    n_layers,
+                    n_heads,
+                    p_dropout,
+                    filter_channels,
+                    mean_only=True,
+                    wn_sharing_parameter=self.wn,
+                    gin_channels=self.gin_channels,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+
+class StochasticDurationPredictor(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        filter_channels = in_channels  # it needs to be removed from future version.
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.log_flow = modules.Log()
+        self.flows = nn.ModuleList()
+        self.flows.append(modules.ElementwiseAffine(2))
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3)
+            )
+            self.flows.append(modules.Flip())
+
+        self.post_pre = nn.Conv1d(1, filter_channels, 1)
+        self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1)
+        self.post_convs = modules.DDSConv(
+            filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout
+        )
+        self.post_flows = nn.ModuleList()
+        self.post_flows.append(modules.ElementwiseAffine(2))
+        for i in range(4):
+            self.post_flows.append(
+                modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3)
+            )
+            self.post_flows.append(modules.Flip())
+
+        self.pre = nn.Conv1d(in_channels, filter_channels, 1)
+        self.proj = nn.Conv1d(filter_channels, filter_channels, 1)
+        self.convs = modules.DDSConv(
+            filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout
+        )
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, filter_channels, 1)
+
+    def forward(self, x, x_mask, w=None, g=None, reverse=False, noise_scale=1.0):
+        x = torch.detach(x)
+        x = self.pre(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.convs(x, x_mask)
+        x = self.proj(x) * x_mask
+
+        if not reverse:
+            flows = self.flows
+            assert w is not None
+
+            logdet_tot_q = 0
+            h_w = self.post_pre(w)
+            h_w = self.post_convs(h_w, x_mask)
+            h_w = self.post_proj(h_w) * x_mask
+            e_q = (
+                torch.randn(w.size(0), 2, w.size(2)).to(device=x.device, dtype=x.dtype)
+                * x_mask
+            )
+            z_q = e_q
+            for flow in self.post_flows:
+                z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w))
+                logdet_tot_q += logdet_q
+            z_u, z1 = torch.split(z_q, [1, 1], 1)
+            u = torch.sigmoid(z_u) * x_mask
+            z0 = (w - u) * x_mask
+            logdet_tot_q += torch.sum(
+                (F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1, 2]
+            )
+            logq = (
+                torch.sum(-0.5 * (math.log(2 * math.pi) + (e_q**2)) * x_mask, [1, 2])
+                - logdet_tot_q
+            )
+
+            logdet_tot = 0
+            z0, logdet = self.log_flow(z0, x_mask)
+            logdet_tot += logdet
+            z = torch.cat([z0, z1], 1)
+            for flow in flows:
+                z, logdet = flow(z, x_mask, g=x, reverse=reverse)
+                logdet_tot = logdet_tot + logdet
+            nll = (
+                torch.sum(0.5 * (math.log(2 * math.pi) + (z**2)) * x_mask, [1, 2])
+                - logdet_tot
+            )
+            return nll + logq  # [b]
+        else:
+            flows = list(reversed(self.flows))
+            flows = flows[:-2] + [flows[-1]]  # remove a useless vflow
+            z = (
+                torch.randn(x.size(0), 2, x.size(2)).to(device=x.device, dtype=x.dtype)
+                * noise_scale
+            )
+            for flow in flows:
+                z = flow(z, x_mask, g=x, reverse=reverse)
+            z0, z1 = torch.split(z, [1, 1], 1)
+            logw = z0
+            return logw
+
+
+class DurationPredictor(nn.Module):
+    def __init__(
+        self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0
+    ):
+        super().__init__()
+
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.gin_channels = gin_channels
+
+        self.drop = nn.Dropout(p_dropout)
+        self.conv_1 = nn.Conv1d(
+            in_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_1 = modules.LayerNorm(filter_channels)
+        self.conv_2 = nn.Conv1d(
+            filter_channels, filter_channels, kernel_size, padding=kernel_size // 2
+        )
+        self.norm_2 = modules.LayerNorm(filter_channels)
+        self.proj = nn.Conv1d(filter_channels, 1, 1)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, in_channels, 1)
+
+    def forward(self, x, x_mask, g=None):
+        x = torch.detach(x)
+        if g is not None:
+            g = torch.detach(g)
+            x = x + self.cond(g)
+        x = self.conv_1(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_1(x)
+        x = self.drop(x)
+        x = self.conv_2(x * x_mask)
+        x = torch.relu(x)
+        x = self.norm_2(x)
+        x = self.drop(x)
+        x = self.proj(x * x_mask)
+        return x * x_mask
+
+
+class TextEncoder(nn.Module):
+    def __init__(
+        self,
+        n_vocab,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        gin_channels=0,
+        num_languages=None,
+        num_tones=None,
+    ):
+        super().__init__()
+        if num_languages is None:
+            from text import num_languages
+        if num_tones is None:
+            from text import num_tones
+        self.n_vocab = n_vocab
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.gin_channels = gin_channels
+        self.emb = nn.Embedding(n_vocab, hidden_channels)
+        nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
+        self.tone_emb = nn.Embedding(num_tones, hidden_channels)
+        nn.init.normal_(self.tone_emb.weight, 0.0, hidden_channels**-0.5)
+        self.language_emb = nn.Embedding(num_languages, hidden_channels)
+        nn.init.normal_(self.language_emb.weight, 0.0, hidden_channels**-0.5)
+        self.bert_proj = nn.Conv1d(1024, hidden_channels, 1)
+        self.ja_bert_proj = nn.Conv1d(768, hidden_channels, 1)
+
+        self.encoder = attentions.Encoder(
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            gin_channels=self.gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, tone, language, bert, ja_bert, g=None):
+        bert_emb = self.bert_proj(bert).transpose(1, 2)
+        ja_bert_emb = self.ja_bert_proj(ja_bert).transpose(1, 2)
+        x = (
+            self.emb(x)
+            + self.tone_emb(tone)
+            + self.language_emb(language)
+            + bert_emb
+            + ja_bert_emb
+        ) * math.sqrt(
+            self.hidden_channels
+        )  # [b, t, h]
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+
+        x = self.encoder(x * x_mask, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return x, m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None, tau=1.0):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * tau * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        print("Removing weight norm...")
+        for layer in self.ups:
+            remove_weight_norm(layer)
+        for layer in self.resblocks:
+            layer.remove_weight_norm()
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm is False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for layer in self.convs:
+            x = layer(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm is False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for layer in self.convs:
+            x = layer(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class ReferenceEncoder(nn.Module):
+    """
+    inputs --- [N, Ty/r, n_mels*r]  mels
+    outputs --- [N, ref_enc_gru_size]
+    """
+
+    def __init__(self, spec_channels, gin_channels=0, layernorm=False):
+        super().__init__()
+        self.spec_channels = spec_channels
+        ref_enc_filters = [32, 32, 64, 64, 128, 128]
+        K = len(ref_enc_filters)
+        filters = [1] + ref_enc_filters
+        convs = [
+            weight_norm(
+                nn.Conv2d(
+                    in_channels=filters[i],
+                    out_channels=filters[i + 1],
+                    kernel_size=(3, 3),
+                    stride=(2, 2),
+                    padding=(1, 1),
+                )
+            )
+            for i in range(K)
+        ]
+        self.convs = nn.ModuleList(convs)
+        # self.wns = nn.ModuleList([weight_norm(num_features=ref_enc_filters[i]) for i in range(K)]) # noqa: E501
+
+        out_channels = self.calculate_channels(spec_channels, 3, 2, 1, K)
+        self.gru = nn.GRU(
+            input_size=ref_enc_filters[-1] * out_channels,
+            hidden_size=256 // 2,
+            batch_first=True,
+        )
+        self.proj = nn.Linear(128, gin_channels)
+        if layernorm:
+            self.layernorm = nn.LayerNorm(self.spec_channels)
+            print('[Ref Enc]: using layer norm')
+        else:
+            self.layernorm = None
+
+    def forward(self, inputs, mask=None):
+        N = inputs.size(0)
+
+        out = inputs.view(N, 1, -1, self.spec_channels)  # [N, 1, Ty, n_freqs]
+        if self.layernorm is not None:
+            out = self.layernorm(out)
+
+        for conv in self.convs:
+            out = conv(out)
+            # out = wn(out)
+            out = F.relu(out)  # [N, 128, Ty//2^K, n_mels//2^K]
+
+        out = out.transpose(1, 2)  # [N, Ty//2^K, 128, n_mels//2^K]
+        T = out.size(1)
+        N = out.size(0)
+        out = out.contiguous().view(N, T, -1)  # [N, Ty//2^K, 128*n_mels//2^K]
+
+        self.gru.flatten_parameters()
+        memory, out = self.gru(out)  # out --- [1, N, 128]
+
+        return self.proj(out.squeeze(0))
+
+    def calculate_channels(self, L, kernel_size, stride, pad, n_convs):
+        for i in range(n_convs):
+            L = (L - kernel_size + 2 * pad) // stride + 1
+        return L
+
+
+class SynthesizerTrn(nn.Module):
+    """
+    Synthesizer for Training
+    """
+
+    def __init__(
+        self,
+        n_vocab,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        n_speakers=256,
+        gin_channels=256,
+        use_sdp=True,
+        n_flow_layer=4,
+        n_layers_trans_flow=6,
+        flow_share_parameter=False,
+        use_transformer_flow=True,
+        use_vc=False,
+        num_languages=None,
+        num_tones=None,
+        norm_refenc=False,
+        **kwargs
+    ):
+        super().__init__()
+        self.n_vocab = n_vocab
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.n_speakers = n_speakers
+        self.gin_channels = gin_channels
+        self.n_layers_trans_flow = n_layers_trans_flow
+        self.use_spk_conditioned_encoder = kwargs.get(
+            "use_spk_conditioned_encoder", True
+        )
+        self.use_sdp = use_sdp
+        self.use_noise_scaled_mas = kwargs.get("use_noise_scaled_mas", False)
+        self.mas_noise_scale_initial = kwargs.get("mas_noise_scale_initial", 0.01)
+        self.noise_scale_delta = kwargs.get("noise_scale_delta", 2e-6)
+        self.current_mas_noise_scale = self.mas_noise_scale_initial
+        if self.use_spk_conditioned_encoder and gin_channels > 0:
+            self.enc_gin_channels = gin_channels
+        else:
+            self.enc_gin_channels = 0
+        self.enc_p = TextEncoder(
+            n_vocab,
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            p_dropout,
+            gin_channels=self.enc_gin_channels,
+            num_languages=num_languages,
+            num_tones=num_tones,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        if use_transformer_flow:
+            self.flow = TransformerCouplingBlock(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers_trans_flow,
+                5,
+                p_dropout,
+                n_flow_layer,
+                gin_channels=gin_channels,
+                share_parameter=flow_share_parameter,
+            )
+        else:
+            self.flow = ResidualCouplingBlock(
+                inter_channels,
+                hidden_channels,
+                5,
+                1,
+                n_flow_layer,
+                gin_channels=gin_channels,
+            )
+        self.sdp = StochasticDurationPredictor(
+            hidden_channels, 192, 3, 0.5, 4, gin_channels=gin_channels
+        )
+        self.dp = DurationPredictor(
+            hidden_channels, 256, 3, 0.5, gin_channels=gin_channels
+        )
+
+        if n_speakers > 0:
+            self.emb_g = nn.Embedding(n_speakers, gin_channels)
+        else:
+            self.ref_enc = ReferenceEncoder(spec_channels, gin_channels, layernorm=norm_refenc)
+        self.use_vc = use_vc
+
+
+    def forward(self, x, x_lengths, y, y_lengths, sid, tone, language, bert, ja_bert):
+        if self.n_speakers > 0:
+            g = self.emb_g(sid).unsqueeze(-1)  # [b, h, 1]
+        else:
+            g = self.ref_enc(y.transpose(1, 2)).unsqueeze(-1)
+        if self.use_vc:
+            g_p = None
+        else:
+            g_p = g
+        x, m_p, logs_p, x_mask = self.enc_p(
+            x, x_lengths, tone, language, bert, ja_bert, g=g_p
+        )
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+
+        with torch.no_grad():
+            # negative cross-entropy
+            s_p_sq_r = torch.exp(-2 * logs_p)  # [b, d, t]
+            neg_cent1 = torch.sum(
+                -0.5 * math.log(2 * math.pi) - logs_p, [1], keepdim=True
+            )  # [b, 1, t_s]
+            neg_cent2 = torch.matmul(
+                -0.5 * (z_p**2).transpose(1, 2), s_p_sq_r
+            )  # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+            neg_cent3 = torch.matmul(
+                z_p.transpose(1, 2), (m_p * s_p_sq_r)
+            )  # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
+            neg_cent4 = torch.sum(
+                -0.5 * (m_p**2) * s_p_sq_r, [1], keepdim=True
+            )  # [b, 1, t_s]
+            neg_cent = neg_cent1 + neg_cent2 + neg_cent3 + neg_cent4
+            if self.use_noise_scaled_mas:
+                epsilon = (
+                    torch.std(neg_cent)
+                    * torch.randn_like(neg_cent)
+                    * self.current_mas_noise_scale
+                )
+                neg_cent = neg_cent + epsilon
+
+            attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+            attn = (
+                monotonic_align.maximum_path(neg_cent, attn_mask.squeeze(1))
+                .unsqueeze(1)
+                .detach()
+            )
+
+        w = attn.sum(2)
+
+        l_length_sdp = self.sdp(x, x_mask, w, g=g)
+        l_length_sdp = l_length_sdp / torch.sum(x_mask)
+
+        logw_ = torch.log(w + 1e-6) * x_mask
+        logw = self.dp(x, x_mask, g=g)
+        l_length_dp = torch.sum((logw - logw_) ** 2, [1, 2]) / torch.sum(
+            x_mask
+        )  # for averaging
+
+        l_length = l_length_dp + l_length_sdp
+
+        # expand prior
+        m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2)
+        logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2)
+
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return (
+            o,
+            l_length,
+            attn,
+            ids_slice,
+            x_mask,
+            y_mask,
+            (z, z_p, m_p, logs_p, m_q, logs_q),
+            (x, logw, logw_),
+        )
+
+    def infer(
+        self,
+        x,
+        x_lengths,
+        sid,
+        tone,
+        language,
+        bert,
+        ja_bert,
+        noise_scale=0.667,
+        length_scale=1,
+        noise_scale_w=0.8,
+        max_len=None,
+        sdp_ratio=0,
+        y=None,
+        g=None,
+    ):
+        # x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths, tone, language, bert)
+        # g = self.gst(y)
+        if g is None:
+            if self.n_speakers > 0:
+                g = self.emb_g(sid).unsqueeze(-1)  # [b, h, 1]
+            else:
+                g = self.ref_enc(y.transpose(1, 2)).unsqueeze(-1)
+        if self.use_vc:
+            g_p = None
+        else:
+            g_p = g
+        x, m_p, logs_p, x_mask = self.enc_p(
+            x, x_lengths, tone, language, bert, ja_bert, g=g_p
+        )
+        logw = self.sdp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w) * (
+            sdp_ratio
+        ) + self.dp(x, x_mask, g=g) * (1 - sdp_ratio)
+        w = torch.exp(logw) * x_mask * length_scale
+        
+        w_ceil = torch.ceil(w)
+        y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
+        y_mask = torch.unsqueeze(commons.sequence_mask(y_lengths, None), 1).to(
+            x_mask.dtype
+        )
+        attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
+        attn = commons.generate_path(w_ceil, attn_mask)
+
+        m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(
+            1, 2
+        )  # [b, t', t], [b, t, d] -> [b, d, t']
+        logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(
+            1, 2
+        )  # [b, t', t], [b, t, d] -> [b, d, t']
+
+        z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
+        z = self.flow(z_p, y_mask, g=g, reverse=True)
+        o = self.dec((z * y_mask)[:, :, :max_len], g=g)
+        # print('max/min of o:', o.max(), o.min())
+        return o, attn, y_mask, (z, z_p, m_p, logs_p)
+
+    def voice_conversion(self, y, y_lengths, sid_src, sid_tgt, tau=1.0):        
+        g_src = sid_src
+        g_tgt = sid_tgt
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src, tau=tau)
+        z_p = self.flow(z, y_mask, g=g_src)
+        z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True)
+        o_hat = self.dec(z_hat * y_mask, g=g_tgt)
+        return o_hat, y_mask, (z, z_p, z_hat)

melo/modules.py

@@ -0,0 +1,598 @@
+import math
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+from . import commons
+from .commons import init_weights, get_padding
+from .transforms import piecewise_rational_quadratic_transform
+from .attentions import Encoder
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        hidden_channels,
+        out_channels,
+        kernel_size,
+        n_layers,
+        p_dropout,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(
+                in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
+            )
+        )
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(
+                    hidden_channels,
+                    hidden_channels,
+                    kernel_size,
+                    padding=kernel_size // 2,
+                )
+            )
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+    Dialted and Depth-Separable Convolution
+    """
+
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+
+        self.drop = nn.Dropout(p_dropout)
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(
+                    channels,
+                    channels,
+                    kernel_size,
+                    groups=channels,
+                    dilation=dilation,
+                    padding=padding,
+                )
+            )
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g=None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+        p_dropout=0,
+    ):
+        super(WN, self).__init__()
+        assert kernel_size % 2 == 1
+        self.hidden_channels = hidden_channels
+        self.kernel_size = (kernel_size,)
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = p_dropout
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(p_dropout)
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(
+                gin_channels, 2 * hidden_channels * n_layers, 1
+            )
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(
+                hidden_channels,
+                2 * hidden_channels,
+                kernel_size,
+                dilation=dilation,
+                padding=padding,
+            )
+            in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(self, x, x_mask, g=None, **kwargs):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i in range(self.n_layers):
+            x_in = self.in_layers[i](x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = self.res_skip_layers[i](acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, : self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels :, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[2],
+                        padding=get_padding(kernel_size, dilation[2]),
+                    )
+                ),
+            ]
+        )
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+            ]
+        )
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+            ]
+        )
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    def forward(self, x, *args, reverse=False, **kwargs):
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        p_dropout=0,
+        gin_channels=0,
+        mean_only=False,
+    ):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            p_dropout=p_dropout,
+            gin_channels=gin_channels,
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+
+class ConvFlow(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        n_layers,
+        num_bins=10,
+        tail_bound=5.0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
+        self.proj = nn.Conv1d(
+            filter_channels, self.half_channels * (num_bins * 3 - 1), 1
+        )
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
+        unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
+            self.filter_channels
+        )
+        unnormalized_derivatives = h[..., 2 * self.num_bins :]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x
+
+
+class TransformerCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        n_layers,
+        n_heads,
+        p_dropout=0,
+        filter_channels=0,
+        mean_only=False,
+        wn_sharing_parameter=None,
+        gin_channels=0,
+    ):
+        assert n_layers == 3, n_layers
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = (
+            Encoder(
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+                isflow=True,
+                gin_channels=gin_channels,
+            )
+            if wn_sharing_parameter is None
+            else wn_sharing_parameter
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x

melo/monotonic_align/__init__.py

@@ -0,0 +1,16 @@
+from numpy import zeros, int32, float32
+from torch import from_numpy
+
+from .core import maximum_path_jit
+
+
+def maximum_path(neg_cent, mask):
+    device = neg_cent.device
+    dtype = neg_cent.dtype
+    neg_cent = neg_cent.data.cpu().numpy().astype(float32)
+    path = zeros(neg_cent.shape, dtype=int32)
+
+    t_t_max = mask.sum(1)[:, 0].data.cpu().numpy().astype(int32)
+    t_s_max = mask.sum(2)[:, 0].data.cpu().numpy().astype(int32)
+    maximum_path_jit(path, neg_cent, t_t_max, t_s_max)
+    return from_numpy(path).to(device=device, dtype=dtype)

melo/monotonic_align/core.py

@@ -0,0 +1,46 @@
+import numba
+
+
+@numba.jit(
+    numba.void(
+        numba.int32[:, :, ::1],
+        numba.float32[:, :, ::1],
+        numba.int32[::1],
+        numba.int32[::1],
+    ),
+    nopython=True,
+    nogil=True,
+)
+def maximum_path_jit(paths, values, t_ys, t_xs):
+    b = paths.shape[0]
+    max_neg_val = -1e9
+    for i in range(int(b)):
+        path = paths[i]
+        value = values[i]
+        t_y = t_ys[i]
+        t_x = t_xs[i]
+
+        v_prev = v_cur = 0.0
+        index = t_x - 1
+
+        for y in range(t_y):
+            for x in range(max(0, t_x + y - t_y), min(t_x, y + 1)):
+                if x == y:
+                    v_cur = max_neg_val
+                else:
+                    v_cur = value[y - 1, x]
+                if x == 0:
+                    if y == 0:
+                        v_prev = 0.0
+                    else:
+                        v_prev = max_neg_val
+                else:
+                    v_prev = value[y - 1, x - 1]
+                value[y, x] += max(v_prev, v_cur)
+
+        for y in range(t_y - 1, -1, -1):
+            path[y, index] = 1
+            if index != 0 and (
+                index == y or value[y - 1, index] < value[y - 1, index - 1]
+            ):
+                index = index - 1

melo/preprocess_text.py

@@ -0,0 +1,135 @@
+import json
+from collections import defaultdict
+from random import shuffle
+from typing import Optional
+
+from tqdm import tqdm
+import click
+from text.cleaner import clean_text_bert
+import os
+import torch
+from text.symbols import symbols, num_languages, num_tones
+
+@click.command()
+@click.option(
+    "--metadata",
+    default="data/example/metadata.list",
+    type=click.Path(exists=True, file_okay=True, dir_okay=False),
+)
+@click.option("--cleaned-path", default=None)
+@click.option("--train-path", default=None)
+@click.option("--val-path", default=None)
+@click.option(
+    "--config_path",
+    default="configs/config.json",
+    type=click.Path(exists=True, file_okay=True, dir_okay=False),
+)
+@click.option("--val-per-spk", default=4)
+@click.option("--max-val-total", default=8)
+@click.option("--clean/--no-clean", default=True)
+def main(
+    metadata: str,
+    cleaned_path: Optional[str],
+    train_path: str,
+    val_path: str,
+    config_path: str,
+    val_per_spk: int,
+    max_val_total: int,
+    clean: bool,
+):
+    if train_path is None:
+        train_path = os.path.join(os.path.dirname(metadata), 'train.list')
+    if val_path is None:
+        val_path = os.path.join(os.path.dirname(metadata), 'val.list')
+    out_config_path = os.path.join(os.path.dirname(metadata), 'config.json')
+
+    if cleaned_path is None:
+        cleaned_path = metadata + ".cleaned"
+
+    if clean:
+        out_file = open(cleaned_path, "w", encoding="utf-8")
+        new_symbols = []
+        for line in tqdm(open(metadata, encoding="utf-8").readlines()):
+            try:
+                utt, spk, language, text = line.strip().split("|")
+                norm_text, phones, tones, word2ph, bert = clean_text_bert(text, language, device='cuda:0')
+                for ph in phones:
+                    if ph not in symbols and ph not in new_symbols:
+                        new_symbols.append(ph)
+                        print('update!, now symbols:')
+                        print(new_symbols)
+                        with open(f'{language}_symbol.txt', 'w') as f:
+                            f.write(f'{new_symbols}')
+
+                assert len(phones) == len(tones)
+                assert len(phones) == sum(word2ph)
+                out_file.write(
+                    "{}|{}|{}|{}|{}|{}|{}\n".format(
+                        utt,
+                        spk,
+                        language,
+                        norm_text,
+                        " ".join(phones),
+                        " ".join([str(i) for i in tones]),
+                        " ".join([str(i) for i in word2ph]),
+                    )
+                )
+                bert_path = utt.replace(".wav", ".bert.pt")
+                os.makedirs(os.path.dirname(bert_path), exist_ok=True)
+                torch.save(bert.cpu(), bert_path)
+            except Exception as error:
+                print("err!", line, error)
+
+        out_file.close()
+
+        metadata = cleaned_path
+
+    spk_utt_map = defaultdict(list)
+    spk_id_map = {}
+    current_sid = 0
+
+    with open(metadata, encoding="utf-8") as f:
+        for line in f.readlines():
+            utt, spk, language, text, phones, tones, word2ph = line.strip().split("|")
+            spk_utt_map[spk].append(line)
+
+            if spk not in spk_id_map.keys():
+                spk_id_map[spk] = current_sid
+                current_sid += 1
+
+    train_list = []
+    val_list = []
+
+    for spk, utts in spk_utt_map.items():
+        shuffle(utts)
+        val_list += utts[:val_per_spk]
+        train_list += utts[val_per_spk:]
+
+    if len(val_list) > max_val_total:
+        train_list += val_list[max_val_total:]
+        val_list = val_list[:max_val_total]
+
+    with open(train_path, "w", encoding="utf-8") as f:
+        for line in train_list:
+            f.write(line)
+
+    with open(val_path, "w", encoding="utf-8") as f:
+        for line in val_list:
+            f.write(line)
+
+    config = json.load(open(config_path, encoding="utf-8"))
+    config["data"]["spk2id"] = spk_id_map
+
+    config["data"]["training_files"] = train_path
+    config["data"]["validation_files"] = val_path
+    config["data"]["n_speakers"] = len(spk_id_map)
+    config["num_languages"] = num_languages
+    config["num_tones"] = num_tones
+    config["symbols"] = symbols
+
+    with open(out_config_path, "w", encoding="utf-8") as f:
+        json.dump(config, f, indent=2, ensure_ascii=False)
+
+
+if __name__ == "__main__":
+    main()

melo/split_utils.py

@@ -0,0 +1,174 @@
+import re
+import os
+import glob
+import numpy as np
+import soundfile as sf
+import torchaudio
+import re
+
+def split_sentence(text, min_len=10, language_str='EN'):
+    if language_str in ['EN', 'FR', 'ES', 'SP']:
+        sentences = split_sentences_latin(text, min_len=min_len)
+    else:
+        sentences = split_sentences_zh(text, min_len=min_len)
+    return sentences
+
+
+def split_sentences_latin(text, min_len=10):
+    text = re.sub('[。！？；]', '.', text)
+    text = re.sub('[，]', ',', text)
+    text = re.sub('[“”]', '"', text)
+    text = re.sub('[‘’]', "'", text)
+    text = re.sub(r"[\<\>\(\)\[\]\"\«\»]+", "", text)
+    return [item.strip() for item in txtsplit(text, 256, 512) if item.strip()]
+
+
+def split_sentences_zh(text, min_len=10):
+    text = re.sub('[。！？；]', '.', text)
+    text = re.sub('[，]', ',', text)
+    # 将文本中的换行符、空格和制表符替换为空格
+    text = re.sub('[\n\t ]+', ' ', text)
+    # 在标点符号后添加一个空格
+    text = re.sub('([,.!?;])', r'\1 $#!', text)
+    # 分隔句子并去除前后空格
+    # sentences = [s.strip() for s in re.split('(。|！|？|；)', text)]
+    sentences = [s.strip() for s in text.split('$#!')]
+    if len(sentences[-1]) == 0: del sentences[-1]
+
+    new_sentences = []
+    new_sent = []
+    count_len = 0
+    for ind, sent in enumerate(sentences):
+        new_sent.append(sent)
+        count_len += len(sent)
+        if count_len > min_len or ind == len(sentences) - 1:
+            count_len = 0
+            new_sentences.append(' '.join(new_sent))
+            new_sent = []
+    return merge_short_sentences_zh(new_sentences)
+
+
+def merge_short_sentences_en(sens):
+    """Avoid short sentences by merging them with the following sentence.
+
+    Args:
+        List[str]: list of input sentences.
+
+    Returns:
+        List[str]: list of output sentences.
+    """
+    sens_out = []
+    for s in sens:
+        # If the previous sentense is too short, merge them with
+        # the current sentence.
+        if len(sens_out) > 0 and len(sens_out[-1].split(" ")) <= 2:
+            sens_out[-1] = sens_out[-1] + " " + s
+        else:
+            sens_out.append(s)
+    try:
+        if len(sens_out[-1].split(" ")) <= 2:
+            sens_out[-2] = sens_out[-2] + " " + sens_out[-1]
+            sens_out.pop(-1)
+    except:
+        pass
+    return sens_out
+
+
+def merge_short_sentences_zh(sens):
+    # return sens
+    """Avoid short sentences by merging them with the following sentence.
+
+    Args:
+        List[str]: list of input sentences.
+
+    Returns:
+        List[str]: list of output sentences.
+    """
+    sens_out = []
+    for s in sens:
+        # If the previous sentense is too short, merge them with
+        # the current sentence.
+        if len(sens_out) > 0 and len(sens_out[-1]) <= 2:
+            sens_out[-1] = sens_out[-1] + " " + s
+        else:
+            sens_out.append(s)
+    try:
+        if len(sens_out[-1]) <= 2:
+            sens_out[-2] = sens_out[-2] + " " + sens_out[-1]
+            sens_out.pop(-1)
+    except:
+        pass
+    return sens_out
+
+
+
+def txtsplit(text, desired_length=100, max_length=200):
+    """Split text it into chunks of a desired length trying to keep sentences intact."""
+    text = re.sub(r'\n\n+', '\n', text)
+    text = re.sub(r'\s+', ' ', text)
+    text = re.sub(r'[""]', '"', text)
+    text = re.sub(r'([,.?!])', r'\1 ', text)
+    text = re.sub(r'\s+', ' ', text)
+    
+    rv = []
+    in_quote = False
+    current = ""
+    split_pos = []
+    pos = -1
+    end_pos = len(text) - 1
+    def seek(delta):
+        nonlocal pos, in_quote, current
+        is_neg = delta < 0
+        for _ in range(abs(delta)):
+            if is_neg:
+                pos -= 1
+                current = current[:-1]
+            else:
+                pos += 1
+                current += text[pos]
+            if text[pos] == '"':
+                in_quote = not in_quote
+        return text[pos]
+    def peek(delta):
+        p = pos + delta
+        return text[p] if p < end_pos and p >= 0 else ""
+    def commit():
+        nonlocal rv, current, split_pos
+        rv.append(current)
+        current = ""
+        split_pos = []
+    while pos < end_pos:
+        c = seek(1)
+        if len(current) >= max_length:
+            if len(split_pos) > 0 and len(current) > (desired_length / 2):
+                d = pos - split_pos[-1]
+                seek(-d)
+            else:
+                while c not in '!?.\n ' and pos > 0 and len(current) > desired_length:
+                    c = seek(-1)
+            commit()
+        elif not in_quote and (c in '!?\n' or (c in '.,' and peek(1) in '\n ')):
+            while pos < len(text) - 1 and len(current) < max_length and peek(1) in '!?.':
+                c = seek(1)
+            split_pos.append(pos)
+            if len(current) >= desired_length:
+                commit()
+        elif in_quote and peek(1) == '"' and peek(2) in '\n ':
+            seek(2)
+            split_pos.append(pos)
+    rv.append(current)
+    rv = [s.strip() for s in rv]
+    rv = [s for s in rv if len(s) > 0 and not re.match(r'^[\s\.,;:!?]*$', s)]
+    return rv
+
+
+if __name__ == '__main__':
+    zh_text = "好的，我来给你讲一个故事吧。从前有一个小姑娘，她叫做小红。小红非常喜欢在森林里玩耍，她经常会和她的小伙伴们一起去探险。有一天，小红和她的小伙伴们走到了森林深处，突然遇到了一只凶猛的野兽。小红的小伙伴们都吓得不敢动弹，但是小红并没有被吓倒，她勇敢地走向野兽，用她的智慧和勇气成功地制服了野兽，保护了她的小伙伴们。从那以后，小红变得更加勇敢和自信，成为了她小伙伴们心中的英雄。"
+    en_text = "I didn’t know what to do. I said please kill her because it would be better than being kidnapped,” Ben, whose surname CNN is not using for security concerns, said on Wednesday. “It’s a nightmare. I said ‘please kill her, don’t take her there.’"
+    sp_text = "¡Claro! ¿En qué tema te gustaría que te hable en español? Puedo proporcionarte información o conversar contigo sobre una amplia variedad de temas, desde cultura y comida hasta viajes y tecnología. ¿Tienes alguna preferencia en particular?"
+    fr_text = "Bien sûr ! En quelle matière voudriez-vous que je vous parle en français ? Je peux vous fournir des informations ou discuter avec vous sur une grande variété de sujets, que ce soit la culture, la nourriture, les voyages ou la technologie. Avez-vous une préférence particulière ?"
+
+    print(split_sentence(zh_text, language_str='ZH'))
+    print(split_sentence(en_text, language_str='EN'))
+    print(split_sentence(sp_text, language_str='SP'))
+    print(split_sentence(fr_text, language_str='FR'))

melo/text/__init__.py

@@ -0,0 +1,35 @@
+from .symbols import *
+
+
+_symbol_to_id = {s: i for i, s in enumerate(symbols)}
+
+
+def cleaned_text_to_sequence(cleaned_text, tones, language, symbol_to_id=None):
+    """Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+    """
+    symbol_to_id_map = symbol_to_id if symbol_to_id else _symbol_to_id
+    phones = [symbol_to_id_map[symbol] for symbol in cleaned_text]
+    tone_start = language_tone_start_map[language]
+    tones = [i + tone_start for i in tones]
+    lang_id = language_id_map[language]
+    lang_ids = [lang_id for i in phones]
+    return phones, tones, lang_ids
+
+
+def get_bert(norm_text, word2ph, language, device):
+    from .chinese_bert import get_bert_feature as zh_bert
+    from .english_bert import get_bert_feature as en_bert
+    from .japanese_bert import get_bert_feature as jp_bert
+    from .chinese_mix import get_bert_feature as zh_mix_en_bert
+    from .spanish_bert import get_bert_feature as sp_bert
+    from .french_bert import get_bert_feature as fr_bert
+    from .korean import get_bert_feature as kr_bert
+
+    lang_bert_func_map = {"ZH": zh_bert, "EN": en_bert, "JP": jp_bert, 'ZH_MIX_EN': zh_mix_en_bert, 
+                          'FR': fr_bert, 'SP': sp_bert, 'ES': sp_bert, "KR": kr_bert}
+    bert = lang_bert_func_map[language](norm_text, word2ph, device)
+    return bert