def wav2mel(wav, hparams=None):
# mel = Audio2Mel().cuda()(src)
# return mel
hparams = hparams or melgan_hparams
wav = np.pad(wav.flatten(), (_pad_len, _pad_len), mode="reflect")
mel = mel_spectrogram(wav, hparams)
mel = mel / 20
return mel
def audio2mel_mellotron(src):
"""
用aukit模块重现生成mel，和mellotron的频谱适应。
:param src:
:return:
"""
wavs = src.cpu().numpy()
mels = []
for wav in wavs:
wav = wav.flatten()[:-1] # 避免生成多一个空帧频谱
mel = mel_spectrogram(wav, default_hparams)
mels.append(mel)
mels = torch.from_numpy(np.array(mels).astype(np.float32))
return mels
def audio2mel_synthesizer(src):
"""
用aukit模块重现生成mel，和synthesizer的频谱适应。
:param src:
:return:
"""
_pad_len = (synthesizer_hparams.n_fft - synthesizer_hparams.hop_size) // 2
wavs = src.cpu().numpy()
mels = []
for wav in wavs:
wav = np.pad(wav.flatten(), (_pad_len, _pad_len), mode="reflect")
mel = mel_spectrogram(wav, synthesizer_hparams)
mel = mel / 20
mels.append(mel)
mels = torch.from_numpy(np.array(mels).astype(np.float32))
return mels
def audio2mel(src):
# print("wav", src.shape)
# mel = Audio2Mel().cuda()(src)
# print("mel", mel.cpu().shape)
# return mel
# wavs = F.pad(src, (p, p), "reflect")
wavs = src.cpu().numpy()
mels = []
for wav in wavs:
wav = np.pad(wav.flatten(), (_pad_len, _pad_len), mode="reflect")
mel = mel_spectrogram(wav, default_hparams)
mels.append(mel)
mels = torch.from_numpy(np.array(mels).astype(np.float32))
return mels
def wavs2mels(indir: Path, outdir: Path):
for fpath in tqdm(indir.glob("*.wav")):
wav = aukit.load_wav(fpath, sr=16000)
wav = np.pad(wav.flatten(), (_pad_len, _pad_len), mode="reflect")
mel = mel_spectrogram(wav, default_hparams)
np.save(outdir.joinpath(fpath.stem + ".npy"), mel, allow_pickle=False)
def melspectrogram(wav, hparams=None):
wav = np.pad(wav.flatten(), (_pad_len, _pad_len), mode="reflect")
mel = mel_spectrogram(wav, melgan_hparams)
mel = mel / 20
return mel