def get_f0(self, audio, sampling_rate=22050, frame_length=1024,
hop_length=256, f0_min=100, f0_max=300, harm_thresh=0.1):
f0, harmonic_rates, argmins, times = compute_yin(
audio, sampling_rate, frame_length, hop_length, f0_min, f0_max,
harm_thresh)
pad = int((frame_length / hop_length) / 2)
f0 = [0.0] * pad + f0 + [0.0] * pad
f0 = np.array(f0, dtype=np.float32)
return f0
def get_f0(self, audio, sampling_rate=22050, frame_length=1024,
hop_length=256, f0_min=100, f0_max=300, harm_thresh=0.1):
f0, harmonic_rates, argmins, times = compute_yin(
audio, sampling_rate, frame_length, hop_length, f0_min, f0_max,
harm_thresh)
pad = int((frame_length / hop_length) / 2)
f0 = [0.0] * pad + f0 + [0.0] * pad
f0 = np.array(f0, dtype=np.float32)
# f0 = f0[:, :melspec.size(1)]
# 用零向量替换F0
# f0 = torch.zeros(1, melspec.shape[1], dtype=torch.float)
# return melspec, f0
return f0
def transform_f0(wav, hparams):
sampling_rate = hparams.sampling_rate
frame_length = hparams.filter_length
hop_length = hparams.hop_length
f0_min = hparams.f0_min
f0_max = hparams.f0_max
harm_thresh = hparams.harm_thresh
f0, harmonic_rates, argmins, times = compute_yin(
wav, sampling_rate, frame_length, hop_length, f0_min, f0_max,
harm_thresh)
pad = int((frame_length / hop_length) / 2)
f0 = [0.0] * pad + f0 + [0.0] * pad
f0 = np.array(f0, dtype=np.float32)
return f0
