Kyutai Labs finally release finetuning code for Moshi - We can now give it any voice we wish!
Posted by JawGBoi@reddit | LocalLLaMA | View on Reddit | 13 comments
Model repo: https://github.com/kyutai-labs/moshi
yukiarimo@reddit
pkmxtw@reddit
Instead of giving it any voice I would rather give the model intelligence.
Foreign-Beginning-49@reddit
Truest burn 🔥 a burn that hurts because it's so true. It was really fun to play with but gave poor gardening advice. I appreciate their work.
silenceimpaired@reddit
Can you use it as a strong text to speech?
Foreign-Beginning-49@reddit
Not that I am aware thete much better options like kokoro or Orpheus.
JadeSerpant@reddit
Lmfao so true.
Aggressive_Escape386@reddit
Does it mean we can fine tune for other languages now?
shakespear94@reddit
I’m a little behind on experimenting with this. Is it just like sesame?
Enough-Meringue4745@reddit
They were so hesitant for so long and now that there’s competition they release it. Fuckin annoying.
FrermitTheKog@reddit
Why didn't they keep improving it? We should have had something as good as Sesame from them by now. Did they run out of money or just lose interest?
Enough-Meringue4745@reddit
They probably did improve it and theyll release it and not provide training for it lol
FrermitTheKog@reddit
Mainly it needs a better brain.
chopders@reddit
Any sample?