Spaces:
Running
Running
| import time | |
| import json | |
| import soundfile as sf | |
| import torch | |
| import torchaudio | |
| import librosa | |
| import yaml | |
| from munch import Munch | |
| from nltk.tokenize import word_tokenize | |
| from models import * | |
| from utils import * | |
| from text_utils import TextCleaner | |
| import phonemizer | |
| from Utils.PLBERT.util import load_plbert | |
| from Modules.diffusion.sampler import DiffusionSampler, ADPM2Sampler, KarrasSchedule | |
| import pandas as pd | |
| # Setup | |
| torch.backends.cudnn.benchmark = False | |
| torch.backends.cudnn.deterministic = True | |
| import random | |
| import tqdm | |
| import argparse | |
| # Load packages | |
| text_cleaner = TextCleaner() | |
| # Mel Spectrogram transformation | |
| to_mel = torchaudio.transforms.MelSpectrogram(n_mels=80, n_fft=2048, win_length=1200, hop_length=300) | |
| mean, std = -4, 4 | |
| # Load models and configurations | |
| device = 'cuda' if torch.cuda.is_available() else 'cpu' | |
| #LibriModel | |
| config = yaml.safe_load(open("Models/LibriTTS/config.yml")) | |
| #LJModel | |
| # config = yaml.safe_load(open("Models/LJSpeech/Models_LJSpeech_config.yml")) | |
| # load pretrained ASR model | |
| ASR_config = config.get('ASR_config', False) | |
| ASR_path = config.get('ASR_path', False) | |
| text_aligner = load_ASR_models(ASR_path, ASR_config) | |
| # load pretrained F0 model | |
| F0_path = config.get('F0_path', False) | |
| pitch_extractor = load_F0_models(F0_path) | |
| # load BERT model | |
| BERT_path = config.get('PLBERT_dir', False) | |
| plbert = load_plbert(BERT_path) | |
| model_params = recursive_munch(config['model_params']) | |
| model = build_model(model_params, text_aligner, pitch_extractor, plbert) | |
| _ = [model[key].eval() for key in model] | |
| _ = [model[key].to(device) for key in model] | |
| #LibriModel | |
| params_whole = torch.load("Models/LibriTTS/epochs_2nd_00020.pth", map_location='cpu') | |
| #LJModel | |
| # params_whole = torch.load("Models/LJSpeech/epoch_2nd_00100.pth", map_location='cpu') | |
| params = params_whole['net'] | |
| for key in model: | |
| if key in params: | |
| print('%s loaded' % key) | |
| try: | |
| model[key].load_state_dict(params[key]) | |
| except: | |
| from collections import OrderedDict | |
| state_dict = params[key] | |
| new_state_dict = OrderedDict() | |
| for k, v in state_dict.items(): | |
| name = k[7:] # remove `module.` | |
| new_state_dict[name] = v | |
| # load params | |
| model[key].load_state_dict(new_state_dict, strict=False) | |
| # except: | |
| # _load(params[key], model[key]) | |
| _ = [model[key].eval() for key in model] | |
| # Load sampler | |
| sampler = DiffusionSampler( | |
| model.diffusion.diffusion, | |
| sampler=ADPM2Sampler(), | |
| sigma_schedule=KarrasSchedule(sigma_min=0.0001, sigma_max=3.0, rho=9.0), # empirical parameters | |
| clamp=False | |
| ) | |
| # Load phonemizer | |
| global_phonemizer = phonemizer.backend.EspeakBackend(language='en-us', preserve_punctuation=True, with_stress=True) | |
| # Preprocessing functions | |
| def length_to_mask(lengths): | |
| "Gets the mask of the max length" | |
| mask = torch.arange(lengths.max()).unsqueeze(0).expand(lengths.shape[0], -1).type_as(lengths) | |
| mask = torch.gt(mask + 1, lengths.unsqueeze(1)) | |
| return mask | |
| def preprocess(wave): | |
| "Turns wave to mel tensor" | |
| wave_tensor = torch.from_numpy(wave).float() | |
| mel_tensor = to_mel(wave_tensor) | |
| mel_tensor = (torch.log(1e-5 + mel_tensor.unsqueeze(0)) - mean) / std | |
| return mel_tensor | |
| def compute_style(path, device, model): | |
| "Computes the style vector for a given audio file" | |
| wave, sr = librosa.load(path, sr=24000) | |
| audio, index = librosa.effects.trim(wave, top_db=30) | |
| if sr != 24000: | |
| audio = librosa.resample(audio, sr, 24000) | |
| mel_tensor = preprocess(audio).to(device) | |
| with torch.no_grad(): | |
| ref_s = model.style_encoder(mel_tensor.unsqueeze(1)) | |
| ref_p = model.predictor_encoder(mel_tensor.unsqueeze(1)) | |
| return torch.cat([ref_s, ref_p], dim=1) | |
| # Inference function | |
| def inference(text, ref_s, model, device, alpha=0.3, beta=0.7, diffusion_steps=10, embedding_scale=1): | |
| # Preprocess text | |
| text = text.replace('"', '') | |
| ps = global_phonemizer.phonemize([text]) | |
| ps = word_tokenize(ps[0]) | |
| ps = ' '.join(ps) | |
| tokens = text_cleaner(ps) | |
| tokens.insert(0, 0) | |
| tokens = torch.LongTensor(tokens).to(device).unsqueeze(0) | |
| max_length = 512 | |
| if len(tokens) > max_length: | |
| tokens = tokens[:max_length] | |
| with torch.no_grad(): | |
| # Process text | |
| input_lengths = torch.LongTensor([tokens.shape[-1]]).to(device) | |
| text_mask = length_to_mask(input_lengths).to(device) | |
| t_en = model.text_encoder(tokens, input_lengths, text_mask) | |
| bert_dur = model.bert(tokens, attention_mask=(~text_mask).int()) | |
| d_en = model.bert_encoder(bert_dur).transpose(-1, -2) | |
| s_pred = sampler(noise = torch.randn((1, 256)).unsqueeze(1).to(device), | |
| embedding=bert_dur, | |
| embedding_scale=embedding_scale, | |
| features=ref_s, # reference from the same speaker as the embedding | |
| num_steps=diffusion_steps).squeeze(1) | |
| s = s_pred[:, 128:] | |
| ref = s_pred[:, :128] | |
| ref = alpha * ref + (1 - alpha) * ref_s[:, :128] | |
| s = beta * s + (1 - beta) * ref_s[:, 128:] | |
| d = model.predictor.text_encoder(d_en, | |
| s, input_lengths, text_mask) | |
| x, _ = model.predictor.lstm(d) | |
| duration = model.predictor.duration_proj(x) | |
| duration = torch.sigmoid(duration).sum(axis=-1) | |
| pred_dur = torch.round(duration.squeeze()).clamp(min=1) | |
| pred_aln_trg = torch.zeros(input_lengths, int(pred_dur.sum().data)) | |
| c_frame = 0 | |
| for i in range(pred_aln_trg.size(0)): | |
| pred_aln_trg[i, c_frame:c_frame + int(pred_dur[i].data)] = 1 | |
| c_frame += int(pred_dur[i].data) | |
| # encode prosody | |
| en = (d.transpose(-1, -2) @ pred_aln_trg.unsqueeze(0).to(device)) | |
| if model_params.decoder.type == "hifigan": | |
| asr_new = torch.zeros_like(en) | |
| asr_new[:, :, 0] = en[:, :, 0] | |
| asr_new[:, :, 1:] = en[:, :, 0:-1] | |
| en = asr_new | |
| F0_pred, N_pred = model.predictor.F0Ntrain(en, s) | |
| asr = (t_en @ pred_aln_trg.unsqueeze(0).to(device)) | |
| if model_params.decoder.type == "hifigan": | |
| asr_new = torch.zeros_like(asr) | |
| asr_new[:, :, 0] = asr[:, :, 0] | |
| asr_new[:, :, 1:] = asr[:, :, 0:-1] | |
| asr = asr_new | |
| out = model.decoder(asr, | |
| F0_pred, N_pred, ref.squeeze().unsqueeze(0)) | |
| # Return synthesized speech | |
| return out.squeeze().cpu().numpy()[..., :-50] | |
| # Function to generate synthetic voices for multiple texts | |
| def generate_synthetic_voice(transcription, ref_s, model, device, index, output_folder="Output", alpha=0.3, beta=0.7, diffusion_steps=10, embedding_scale=1): | |
| wav = inference(transcription, ref_s, model, device, alpha, beta, diffusion_steps, embedding_scale) | |
| audio_filename = f"synthetic_voice_{index}.wav" | |
| output_path = f"{output_folder}/{audio_filename}" | |
| sf.write(output_path, wav, 24000) | |
| return output_path | |
| def main(max_files): | |
| # Load the transcriptions from the CSV file | |
| csv_file = 'transcriptions.csv' | |
| if os.path.exists(csv_file): | |
| df = pd.read_csv(csv_file) | |
| # Ensure the "audio_path" column exists | |
| if 'audio_path' not in df.columns: | |
| df['audio_path'] = '' | |
| # Get the list of reference audio files | |
| ref_audio_folder = 'reference_audio' | |
| ref_audio_files = [os.path.join(ref_audio_folder, f) for f in os.listdir(ref_audio_folder) if f.endswith('.wav')] | |
| # Calculate the number of rows to process | |
| num_rows_to_process = df['audio_path'].isnull().sum() + (df['audio_path'] == '').sum() | |
| num_rows_to_process = min(num_rows_to_process, max_files) | |
| if max_files == np.inf: | |
| max_files = num_rows_to_process | |
| progress_bar = tqdm.tqdm(total=num_rows_to_process, desc="Generating Audio Files") | |
| new_audio_count = 0 # Initialize the counter for new audio files | |
| # Process transcriptions that don't have an audio path yet | |
| for index, row in df.iterrows(): | |
| if pd.isna(row['audio_path']) or row['audio_path'] == '': | |
| path = random.choice(ref_audio_files) | |
| ref_s = compute_style(path, device, model) | |
| transcription = row['transcription'] | |
| audio_path = generate_synthetic_voice(transcription, ref_s, model, device, index) | |
| df.at[index, 'audio_path'] = audio_path # Write audio path in dataframe | |
| new_audio_count += 1 | |
| progress_bar.update(1) | |
| # Save the updated DataFrame to the CSV file | |
| df.to_csv(csv_file, index=False) | |
| # Stop if the maximum number of new audio files is reached | |
| if new_audio_count >= max_files: | |
| break | |
| if new_audio_count == num_rows_to_process: | |
| print('All transcriptions have been created.') | |
| else: | |
| print(f'Finished creating {new_audio_count} new transcriptions.') | |
| else: | |
| print(f"CSV file {csv_file} does not exist.") | |
| if __name__ == "__main__": | |
| parser = argparse.ArgumentParser(description='Generate synthetic audio files.') | |
| parser.add_argument('--max_files', type=str, default='5', help='Maximum number of new audio files to create or "all" to process all rows') | |
| args = parser.parse_args() | |
| if args.max_files.lower() == 'all': | |
| max_files = np.inf # Set to infinity to process all rows | |
| else: | |
| max_files = int(args.max_files) | |
| main(max_files=max_files) | |
| # Example: python inference.py --max_files=100 | |
| # This will create 100 new audio files or all the transcriptions from the csv file | |