Spaces:

TTS-AGI

/

TTS-Arena

like 496

Running on CPU Upgrade

Seems to be a known issue within the Coqui Discord and said to occur 80% of the time even on TTS Arena. So no matter how well XTTS says a sentence, if the other candidate says it right without a cutoff, then I have to go with that one.

Yeah, I just use inference streaming instead for inferencing. Just concatenate the chunks together in the end. That's exactly what is done in the example they provided in their docs.
A bit slower, but solves most of the problems

User probably referring to these docs: https://github.com/coqui-ai/TTS/blob/dev/docs/source/models/xtts.md#streaming-manually