intunist / nnsvs-english-support Goto Github PK

Hi, I have several questions, could we connect?

Hello,

I found a couple of issues:

The acoustic model doesn't seem to train with the current version of nnsvs. Time lag and duration model work fine, but the loss on the acoustic model doesn't decrease. I tried both my own custom English voicebank and one of the standard Japanese ones - same result. Is that a known issue?

On stage 7, the system expects a myconfig.yaml in ETK/train/conf/train_postfilter, which isn't included in the repo. Can you help?

I get quite a few warnings that the WAV file is shorter than the UST/LAB file. Sometimes this is because the system wants me to add a pause to the end of every UST file even though there's not always a pause at the end of the WAV file. Where UST, LAB and WAV all end on a phoneme, adding a small pause to the end of the UST and LAB files will necessarily result in the WAV file being a bit shorter than the UST and LAB files. Is that a problem?

Related to this is the following problem - sometimes when I get the warnings that the files don't align, the system seems to be comparing the wrong files (see below). Why does this happen?

WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_A_12.wav) (data/full_align_round/jessica_train_song_A_13.lab)
WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_A_13.wav) (data/full_align_round/jessica_train_song_A_14.lab)
WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_A_15.wav) (data/full_align_round/jessica_train_song_B_0.lab)
WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_B_1.wav) (data/full_align_round/jessica_train_song_B_2.lab)
WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_B_4.wav) (data/full_align_round/jessica_train_song_B_5.lab)
WARNING:root:WAV is shorter than LAB or score. (data/wav/jessica_train_song_B_5.wav) (data/full_align_round/jessica_train_song_B_6.lab)

Thank you!

Hello,
I've been to many places to implement new language support for NNSVS. And I found this repo.
nnsvs/nnsvs#63

I think NNSVS can support a new language with only a table and hed file for the new language.

To me, you seem to be an expert.
Do you have detailed instructions to support the new language?

Thank you in advance.