I fixed my tiny TTS by editing the weights directly, no retraining
I'm building Silencio-2, a small finetune of Kyutai's Pocket TTS. It adds inline tags for emotions ( , , , , , ) and for sounds like [laugh] . I simply didn't have enough tagged data to make them stable. worked, the others were hit or miss, and gave an actual whisper in about 1 of 5 takes.