The parameter "frame_length" in compute_f0 should be the "win_length" rather than the "filter_length".
When filter_length > win_length, the f0 computation will be incorrect.
As in mel computation, we first get the window with win_length and then pad it to filter_length.
In F0 computation, we should as well compute F0 within a win_length frame rather than filter_length frame.
candlewill
commented
4 years ago
The parameter "frame_length" in compute_f0 should be the "win_length" rather than the "filter_length". When filter_length > win_length, the f0 computation will be incorrect. As in mel computation, we first get the window with win_length and then pad it to filter_length. In F0 computation, we should as well compute F0 within a win_length frame rather than filter_length frame.