When I re-synthesize singing voice in Pyworld, singing voice is grazed

[Otonari1224]

When I re-synthesize a singing voice in Pyworld, the voice is grazed as shown in the picture. Is there a workaround for this?

[sunnnnnnnny]

@Otonari1224 can you solve it ? I meet same problem

[adityuhkapoor]

Ran into this too

Granular-resynthesized vocals lost their character on a no-op round-trip and running aperiodicity didn't help. seems likely that F0 estimation failing on inharmonic input, cascading into bad spectral envelope. Ended up using PSOLA instead.