Spaces:
Running
on
Zero
Running
on
Zero
Better Voice for FishSpeech
#25
by
ecyht2
- opened
I notice that FishSpeech v1.5 performs better in the original arena compared to the spaces. I know the results are mixed with the previous versions but, I think another difference in the voice given to the model. Perhaps some model's performance might improve if given a better voice. (I wonder what voice they use for the original arena).
I wonder what original Arena uses too. Hence I made this Arena way more open and transparent. And easier to duplicate and build upon.
But overall FS performs better there because the random sentences are narration based. Yes, the FS space switched the model from 1.4 to 1.5 in December. But the felt improvement was mild when I compares the win_rate:
#20