i am confused about this tool about opensvip

[yqzhishen]

Opencpop contains raw wav files so you can simply use pitch extraction algorithms to get the f0

[lvZic]

Opencpop contains raw wav files so you can simply use pitch extraction algorithms to get the f0

thanks for reply.
So the wav file of oencpop used to extract F0 is the human voice? not the melody. I should use the human voice to get F0. is it right?

[yqzhishen]

yes

[lvZic]

yes

i used librosa to get F0:
y, sr = librosa.load(filename, sr=None)
f0_new = librosa.yin(y[:], fmin=librosa.note_to_hz('A1'), fmax=librosa.note_to_hz('G7'), sr=sr)
f0_new[np.isnan(f0_new)] = 0

param['f0_seq'] = f0_new
param['f0_timestamp'] = str(2048 * 1.0 / sr)

however, the final wav file sounds weird, which means the F0 is not fit.

[yqzhishen]

See https://github.com/openvpi/DiffSinger/blob/d79ce3fe8ac8a0832202ce8fe1e48cce17e64aa6/utils/binarizer_utils.py#L12 and be careful about f0_timestep.

I will close this issue since it's not related to this project.