←back to thread

VibeVoice: A Frontier Open-Source Text-to-Speech Model

(microsoft.github.io)
448 points lastdong | 8 comments | 03 Sep 25 10:44 UTC | HN request time: 1.105s | source | bottom
1. aargh_aargh ◴[03 Sep 25 12:34 UTC] No.45114957[source]▶
Is there a current, updated list (ideally, a ranking) of the best open weights TTS models?

I'm actually more interested in STT (ASR) but the choices there are rather limited.

replies(4): >>45115421 #>>45116346 #>>45116488 #>>45116531 #
2. xnx ◴[03 Sep 25 13:19 UTC] No.45115421[source]▶
Click leaderboard in the hamburger menu: https://huggingface.co/spaces/TTS-AGI/TTS-Arena-V2
replies(2): >>45116081 #>>45116291 #
3. prophesi ◴[03 Sep 25 14:15 UTC] No.45116081[source]▶
Is there a way to filter out hosted models? The top three winners currently are all proprietary as far as I can tell.

edit: Ah, there's a lock icon next to the name of each proprietary model.

4. odie5533 ◴[03 Sep 25 14:32 UTC] No.45116291[source]▶
That's a highly incomplete comparison
5. odie5533 ◴[03 Sep 25 14:36 UTC] No.45116346[source]▶
Best TTS: VibeVoice, Chatterbox, Dia, Higgs, F5 TTS, Kokoro, Cosy Voice, XTTS-2.
replies(1): >>45124463 #
6. Uehreka ◴[03 Sep 25 14:50 UTC] No.45116488[source]▶
Yes: https://huggingface.co/models?pipeline_tag=text-to-speech

Generally if a model is trending on that page, there’s enough juice for it to be worth a try. There’s a lot of subjective-opinion-having in this space, so beyond “is it trending on HF” the best eval is your own ears. But if something is not trending on HF it is unlikely to be much good.

7. watsonmusic ◴[03 Sep 25 14:53 UTC] No.45116531[source]▶
yes the best
8. kroaton ◴[04 Sep 25 07:12 UTC] No.45124463[source]▶
Unmute.sh (same team as Kokoro) gets slept on, but it's really good.