Spaces:
Sleeping
Sleeping
| import argparse | |
| import glob | |
| import os | |
| import numpy as np | |
| import torch | |
| import tqdm | |
| from scipy.io.wavfile import write | |
| from .denoiser import Denoiser | |
| from .model.generator import ModifiedGenerator | |
| from .utils.hparams import HParam, load_hparam_str | |
| MAX_WAV_VALUE = 32768.0 | |
| from .download_utils import download_url | |
| url = 'https://zenodo.org/record/4743731/files/vctk_pretrained_model_3180.pt' | |
| class VocGan: | |
| def __init__(self, device='cuda:0',config=None, denoise=False): | |
| home = os.environ['HOME'] | |
| checkpoint_path = os.path.join(home,'./.cache/vocgan') | |
| os.makedirs(checkpoint_path,exist_ok=True) | |
| checkpoint_file = os.path.join(checkpoint_path,'vctk_pretrained_model_3180.pt') | |
| if not os.path.exists(checkpoint_file): | |
| download_url(url,checkpoint_path) | |
| checkpoint = torch.load(checkpoint_file,map_location=device) | |
| if config is not None: | |
| hp = HParam(config) | |
| else: | |
| hp = load_hparam_str(checkpoint['hp_str']) | |
| self.hp = hp | |
| self.model = ModifiedGenerator(hp.audio.n_mel_channels, | |
| hp.model.n_residual_layers, | |
| ratios=hp.model.generator_ratio, | |
| mult=hp.model.mult, | |
| out_band=hp.model.out_channels).to(device) | |
| self.model.load_state_dict(checkpoint['model_g']) | |
| self.model.eval(inference=True) | |
| self.model = self.model.to(device) | |
| self.denoise = denoise | |
| self.device = device | |
| def synthesize(self, mel): | |
| with torch.no_grad(): | |
| if not isinstance(mel,torch.Tensor): | |
| mel = torch.tensor(mel) | |
| if len(mel.shape) == 2: | |
| mel = mel.unsqueeze(0) | |
| mel = mel.to(self.device) | |
| audio = self.model.inference(mel) | |
| audio = audio.squeeze(0) # collapse all dimension except time axis | |
| if self.denoise: | |
| denoiser = Denoiser(self.model,device=self.device) | |
| #.to(self.device) | |
| audio = denoiser(audio, 0.01) | |
| audio = audio.squeeze() | |
| audio = audio[:-(self.hp.audio.hop_length * 10)] | |
| #audio = MAX_WAV_VALUE * audio | |
| #audio = audio.clamp(min=-MAX_WAV_VALUE, max=MAX_WAV_VALUE - 1) | |
| #audio = audio.short() | |
| audio = audio.cpu().detach().numpy() | |
| return audio | |
| def __call__(self,mel): | |
| return self.synthesize(mel) | |