def inference_from_path(self): src_mel, _ = get_spectrograms(self.args.source) tar_mel, _ = get_spectrograms(self.args.target) src_mel = torch.from_numpy(self.normalize(src_mel)).cuda() tar_mel = torch.from_numpy(self.normalize(tar_mel)).cuda() conv_wav, conv_mel = self.inference_one_utterance(src_mel, tar_mel) self.write_wav_to_file(conv_wav, self.args.output) return
def convert_from_path(self, srcpath, trgpath, opath, device_id): #print(f"Running on device {device_id}") os.environ["CUDA_VISIBLE_DEVICES"] = str(device_id) src_mel, _ = get_spectrograms(srcpath) tar_mel, _ = get_spectrograms(trgpath) src_mel = torch.from_numpy(self.normalize(src_mel)).cuda() tar_mel = torch.from_numpy(self.normalize(tar_mel)).cuda() conv_wav, conv_mel = self.convert_one_utterance(src_mel, tar_mel) self.write_wav_to_file(conv_wav, opath) return
def inference_form_path_multi_target(self, target_path_list, save=False): src_mel, _ = get_spectrograms(self.args.source) src_mel = torch.from_numpy(self.normalize(src_mel)) tar_mel_list = [] for target_path in target_path_list: tar_mel, _ = get_spectrograms(target_path) tar_mel = torch.from_numpy(self.normalize(tar_mel)) tar_mel_list.append(tar_mel) conv_wav, conv_mel = self.inference_one_utterance_multi_target( src_mel, tar_mel_list) if save: self.write_wav_to_file(conv_wav, self.args.output) return conv_wav
def inference_one_utterance_spk(self, uttpath): x, _ = get_spectrograms(uttpath) x = torch.from_numpy(self.normalize(x)).cuda() x = self.utt_make_frames(x) emb = self.model.get_speaker_embeddings(x) emb = emb.detach().cpu().numpy() return emb
def inference_one_utterance_content(self, uttpath): x, _ = get_spectrograms(uttpath) x = torch.from_numpy(self.normalize(x)).cuda() x = self.utt_make_frames(x) emb = self.model.get_content_repr(x) emb = emb.detach().cpu().numpy() emb = np.mean(emb, axis=2) return emb
def get_mel_frames(self, apath): tar_mel_frames, _ = get_spectrograms(apath) tar_mel_frames = torch.from_numpy( self.normalize(tar_mel_frames)).cuda() tar_mel_frames = self.utt_make_frames(tar_mel_frames) return tar_mel_frames