Spaces:
Sleeping
Sleeping
File size: 6,601 Bytes
aa7cb02 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 |
import os
import numpy as np
import traceback
from TTS.demos.xtts_ft_demo.utils.gpt_train import train_gpt
import torch
import torchaudio
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
def load_fine_tuned_xtts_v2(config_path, checkpoint_path, reference_audio_path):
"""
Load the fine-tuned XTTS v2 model and compute speaker latents.
Args:
config_path (str): Path to the configuration file.
Example: "path/to/config.json"
checkpoint_path (str): Path to the checkpoint directory.
Example: "path/to/checkpoint/"
reference_audio_path (str): Path to the reference audio file.
Example: "path/to/reference.wav"
Returns:
tuple: A tuple containing the model, gpt_cond_latent, and speaker_embedding.
Example: (model, gpt_cond_latent, speaker_embedding)
"""
print("Loading model...")
config = XttsConfig()
config.load_json(config_path)
model = Xtts.init_from_config(config)
model.load_checkpoint(config, checkpoint_dir=checkpoint_path, use_deepspeed=True)
model.cuda()
print("Computing speaker latents...")
gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=[reference_audio_path])
return model, gpt_cond_latent, speaker_embedding
def Inference(model, gpt_cond_latent, speaker_embedding,path_to_save,text, temperature=0.7):
"""
Perform inference using the fine-tuned XTTS v2 model.
Args:
model (Xtts): The XTTS v2 model.
Example: model, gpt_cond_latent, speaker_embedding = load_fine_tuned_xtts_v2(config_path, checkpoint_path, reference_audio_path)
gpt_cond_latent (torch.Tensor): GPT conditioning latent vectors.
speaker_embedding (torch.Tensor): Speaker embedding vectors.
path_to_save (str): Path to save the generated audio.
Example: "path/to/output.wav"
text (str): The input text for synthesis.
Example: "Hello, world!"
temperature (float, optional): Sampling temperature. Default is 0.7.
Example: 0.7
Returns:
None
"""
print("Inference...")
out = model.inference(
text,
gpt_cond_latent,
speaker_embedding,
temperature, # Add custom parameters here # 3
)
torchaudio.save(path_to_save, torch.tensor(out["wav"]).unsqueeze(0), 24000)
#model, gpt_cond_latent, speaker_embedding = load_fine_tuned_xtts_v2("C:/tmp/xtts_ft/run/training/GPT_XTTS_FT-April-02-2024_05+08PM-0000000/config.json", "C:/tmp/xtts_ft/run/training/GPT_XTTS_FT-April-02-2024_05+08PM-0000000/best_model_72.pth", "old_man_segments/wavs/segment_10.wav")
class xtts_v2_Model():
"""
A class to handle training of the XTTS v2 model.
Args:
train_csv_path (str): Path to the training CSV file.
Example: "path/to/train.csv"
eval_csv_path (str): Path to the evaluation CSV file.
Example: "path/to/eval.csv"
num_epochs (int): Number of training epochs.
Example: 10
batch_size (int): Size of each training batch.
Example: 4
grad_acumm (int): Gradient accumulation steps.
Example: 1
output_path (str): Path to save the trained model outputs.
Example: "path/to/output/"
max_audio_length (int): Maximum allowed length of audio for training in seconds.
Example: 10
language (str, optional): Language of the audio files, either 'en' for English or 'es' for Spanish. Default is "en".
Example: "en"
"""
def __init__(self, train_csv_path, eval_csv_path, num_epochs, batch_size, grad_acumm, output_path, max_audio_length, language="en"):
self.train_csv_path = train_csv_path
self.eval_csv_path = eval_csv_path
self.num_epochs = num_epochs
self.batch_size = batch_size
self.grad_acumm = grad_acumm
self.output_path = output_path
self.max_audio_length = max_audio_length
self.language = language
self.config_path = None
self.original_xtts_checkpoint = None
self.vocab_file = None
self.exp_path = None
self.speaker_wav = None
def train_model(self):
"""
Train the XTTS v2 model.
Returns:
tuple: A tuple containing a status message, config_path, vocab_file, fine-tuned XTTS checkpoint, and speaker wav file.
Example: ("Model training done!", "path/to/config.json", "path/to/vocab.json", "path/to/best_model.pth", "path/to/speaker.wav")
"""
#clear_gpu_cache()
if not self.train_csv_path or not self.eval_csv_path:
return "You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !", "", "", "", ""
try:
# convert seconds to waveform frames
max_audio_length = int(max_audio_length * 22050)
self.config_path, self.original_xtts_checkpoint, self.vocab_file, self.exp_path, self.speaker_wav = train_gpt(self.language, self.num_epochs, self.batch_size, self.grad_acumm, self.train_csv_path, self.eval_csv_path, output_path=self.output_path, max_audio_length=max_audio_length)
except:
traceback.print_exc()
error = traceback.format_exc()
return f"The training was interrupted due an error !! Please check the console to check the full error message! \n Error summary: {error}", "", "", "", ""
# copy original files to avoid parameters changes issues
os.system(f"cp {self.config_path} {self.exp_path}")
os.system(f"cp {self.vocab_file} {self.exp_path}")
ft_xtts_checkpoint = os.path.join(self.exp_path, "best_model.pth")
print("Model training done!")
#clear_gpu_cache()
return "Model training done!", self.config_path, self.vocab_file, ft_xtts_checkpoint, self.speaker_wav
# example
#train_meta = "C:/tmp/xtts_ft/run/training/GPT_XTTS_FT-April-02-2024_05+08PM-0000000/train.csv"
#eval_meta = "C:/tmp/xtts_ft/run/training/GPT_XTTS_FT-April-02-2024_05+08PM-0000000/eval.csv"
#num_epochs = 10
#batch_size = 4
#grad_acumm = 1
#out_path = "C:/tmp/xtts_ft/run/training/GPT_XTTS_FT-April-02-2024_05+08PM-0000000"
#max_audio_length = 10
#lang = "en"
#xtts_v2 = xtts_v2_Model(train_meta, eval_meta, num_epochs, batch_size, grad_acumm, out_path, max_audio_length, lang)
#_, config_path, vocab_path, ft_xtts_checkpoint, speaker_wav = xtts_v2_Model.train_model() |