import re import os import sys import utils import torch import random import commons import numpy as np import gradio as gr from tqdm import tqdm from models import SynthesizerTrn from huggingface_hub import snapshot_download from text import cleaned_text_to_sequence, get_bert from text.cleaner import clean_text from text.symbols import symbols if sys.platform == "darwin": os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1" import logging logging.getLogger("numba").setLevel(logging.WARNING) logging.getLogger("markdown_it").setLevel(logging.WARNING) logging.getLogger("urllib3").setLevel(logging.WARNING) logging.getLogger("matplotlib").setLevel(logging.WARNING) logging.basicConfig( level=logging.INFO, format="| %(name)s | %(levelname)s | %(message)s" ) logger = logging.getLogger(__name__) net_g = None debug = False def get_text(text, language_str, hps): norm_text, phone, tone, word2ph = clean_text(text, language_str) phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str) if hps.data.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 = get_bert(norm_text, word2ph, language_str) del word2ph assert bert.shape[-1] == len(phone) phone = torch.LongTensor(phone) tone = torch.LongTensor(tone) language = torch.LongTensor(language) return bert, phone, tone, language def infer(text, sdp_ratio, noise_scale, noise_scale_w, length_scale, sid): global net_g bert, phones, tones, lang_ids = get_text(text, "ZH", hps) 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) x_tst_lengths = torch.LongTensor([phones.size(0)]).to(device) del phones speakers = torch.LongTensor([hps.data.spk2id[sid]]).to(device) audio = ( net_g.infer( x_tst, x_tst_lengths, speakers, tones, lang_ids, bert, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale, )[0][0, 0] .data.cpu() .float() .numpy() ) del x_tst, tones, lang_ids, bert, x_tst_lengths, speakers return audio def tts_fn(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale): with torch.no_grad(): audio = infer( text, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale, sid=speaker, ) return (hps.data.sampling_rate, audio) def text_splitter(text: str): punctuation = r"[。,；,！,？,〜,\n,\r,\t,.,!,;,?,~, ]" sentences = re.split(punctuation, text.strip()) return [sentence.strip() for sentence in sentences if sentence.strip()] def concatenate_audios(audio_samples, sample_rate=44100): half_second_silence = np.zeros(int(sample_rate / 2)) final_audio = audio_samples[0] for sample in audio_samples[1:]: final_audio = np.concatenate((final_audio, half_second_silence, sample)) print("Audio pieces concatenated!") return (sample_rate, final_audio) def read_text(file_path: str): try: with open(file_path, "r", encoding="utf-8") as file: content = file.read() return content except FileNotFoundError: print(f"File Not Found: {file_path}") except IOError: print(f"An error occurred reading the file: {file_path}") except Exception as e: print(f"An unknown error has occurred: {e}") def infer_tab1(text, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale): try: content = read_text(text) sentences = text_splitter(content) audios = [] for sentence in tqdm(sentences, desc="TTS inferring..."): with torch.no_grad(): audios.append( infer( sentence, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale, sid=speaker, ) ) return concatenate_audios(audios, hps.data.sampling_rate), content except Exception as e: return None, f"{e}" def infer_tab2(content, speaker, sdp_ratio, noise_scale, noise_scale_w, length_scale): try: sentences = text_splitter(content) audios = [] for sentence in tqdm(sentences, desc="TTS inferring..."): with torch.no_grad(): audios.append( infer( sentence, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale, sid=speaker, ) ) return concatenate_audios(audios, hps.data.sampling_rate) except Exception as e: print(f"{e}") return None if __name__ == "__main__": model_dir = snapshot_download("Genius-Society/hoyoTTS", cache_dir="./__pycache__") if debug: logger.info("Enable DEBUG-LEVEL log") logging.basicConfig(level=logging.DEBUG) hps = utils.get_hparams_from_dir(model_dir) device = ( "cuda:0" if torch.cuda.is_available() else ( "mps" if sys.platform == "darwin" and torch.backends.mps.is_available() else "cpu" ) ) net_g = SynthesizerTrn( len(symbols), hps.data.filter_length // 2 + 1, hps.train.segment_size // hps.data.hop_length, n_speakers=hps.data.n_speakers, **hps.model, ).to(device) net_g.eval() utils.load_checkpoint(f"{model_dir}/G_78000.pth", net_g, None, skip_optimizer=True) speaker_ids = hps.data.spk2id speakers = list(speaker_ids.keys()) random.shuffle(speakers) with gr.Blocks() as app: gr.Markdown( """ Welcome to the Space, which is based on the open source project Bert-vits2, and moved to the bottom for an explanation of the principle. This Space must be used in accordance with local laws and regulations, prohibiting the use of it for any criminal activities.""" ) with gr.Tab("Input Mode"): gr.Interface( fn=infer_tab2, inputs=[ gr.TextArea( label="Please input the Simplified Chinese text", placeholder="The first inference takes time to download the model, so be patient.", show_copy_button=True, ), gr.Dropdown(choices=speakers, value="莱依拉", label="Role"), gr.Slider( minimum=0, maximum=1, value=0.2, step=0.1, label="Modulation of intonation", ), # SDP/DP Mix Ratio gr.Slider( minimum=0.1, maximum=2, value=0.6, step=0.1, label="Emotional adjustment", ), gr.Slider( minimum=0.1, maximum=2, value=0.8, step=0.1, label="Phoneme length", ), gr.Slider( minimum=0.1, maximum=2, value=1, step=0.1, label="Output duration", ), ], outputs=gr.Audio(label="Output Audio", show_share_button=False), flagging_mode="never", concurrency_limit=4, ) with gr.Tab("Upload Mode"): gr.Interface( fn=infer_tab1, # Use text_to_speech func inputs=[ gr.components.File( label="Please upload a simplified Chinese TXT", type="filepath", file_types=[".txt"], ), gr.Dropdown(choices=speakers, value="莱依拉", label="Role"), gr.Slider( minimum=0, maximum=1, value=0.2, step=0.1, label="Modulation of intonation", ), gr.Slider( minimum=0.1, maximum=2, value=0.6, step=0.1, label="Emotional adjustment", ), gr.Slider( minimum=0.1, maximum=2, value=0.8, step=0.1, label="Phoneme length", ), gr.Slider( minimum=0.1, maximum=2, value=1, step=0.1, label="Output duration", ), ], outputs=[ gr.Audio(label="Output Audio", show_share_button=False), gr.TextArea( label="Result of TXT extraction", show_copy_button=True, ), ], flagging_mode="never", concurrency_limit=4, ) gr.HTML( """ """ ) app.launch()