Converting Enunu to NNSVS

[Surgbc]

I have used the provided script to convert some Enunu model to the format required by NNSVS. But not all models work in NNSVS, giving only silent audio.

For instance the trainer of AICHI has told me that: a.i.chi was trained on the ENUNU training kit so he will *only work alongside the plug-in!* Should this be the case?

[r9y9]

Could you describe detailed steps to reproduce? I believe it should work merely with NNSVS.

[Surgbc]

I have created a notebook:

It has a comparison also of the outputs for the Ritsu model which is also converted from ENUNU and Yoko for both a Japanese song and an English song.

[r9y9]

It looks like something is wrong with the full-context extraction.

  contexts = pysinsy.extract_fullcontext(file_)
  labels = hts.HTSLabelFile.create_from_contexts(contexts)

This part only works correct if the full-context labels for training are consistent with Sinsy's format. For Ritsu, you'd need to use https://github.com/oatsu-gh/utaupy to get full-context labels from UST. I'm not sure about a.i.chi.