piper icon indicating copy to clipboard operation
piper copied to clipboard

Kurdish model needed

Open willwade opened this issue 1 year ago • 3 comments

We desperately need Kurdish TTS to support people without speech. There are no TTS systems supporting this (apart from eSpeak)

I'm really parking this here as a note to anyone else wanting to do this too. Here are some datasets

  • https://github.com/AsoSoft/AsoSoft-TTS-Speech-Corpus-for-Central-Kurdish
  • https://data.mendeley.com/datasets/gft65z43hs/1

(sort of related:

  • https://huggingface.co/Akashpb13/Central_kurdish_xlsr
  • https://huggingface.co/datasets/navinaananthan/Kurdish-Sorani-Parallel-Corpus )

willwade avatar May 14 '24 06:05 willwade

The next version of Piper will be moving away from espeak (for licensing reasons), so I will need a pronunciation dictionary for Kurdish similar to these: https://mfa-models.readthedocs.io/en/latest/dictionary/index.html#dictionary

The dictionary can be generated with espeak-ng too, though the quality depends on aspects of the language itself. Arabic is especially difficult because the written form can be spoken many different ways depending on context.

synesthesiam avatar May 18 '24 17:05 synesthesiam

Interesting. Any info on the licensing issues with espeak??

willwade avatar May 18 '24 21:05 willwade

espeak and its successor espeak-ng are LGPL. Depending on who you ask, this means that any project using it should also be some form of GPL. Since Piper is MIT licensed, I want to make sure there's no question that it can be used everywhere.

synesthesiam avatar May 18 '24 22:05 synesthesiam