the synthesized voice trembles

[xiaoyangnihao]

@MlWoo Hi, I use LPCNet as vocoder to train female and male dataset seperately. The synthesized voice of female is OK, but the synthesized voice of male trembles. Have you encountered this problem? Do I need or how to modify the part of feature extraction?

[MlWoo]

@xiaoyangnihao usually related to the pitch range and bank filters.

[xiaoyangnihao]

@xiaoyangnihao usually related to the pitch range and bank filters.

Thanks for your quick reply, So I want to ask

1. pitch range: The fundamental frequency of male is lower than that of female, so the PITCH_MIN_PERIOD和PITCH_MAX_PERIOD should be increased？
2. bank filters：how should I understand this point ? I think 18 bands are enough. By the way, can I have your wechat ID if possible, thanks.