←back to thread

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

(microsoft.github.io)
448 points lastdong | 1 comments | 03 Sep 25 10:44 UTC | HN request time: 0.327s | source
Show context
baal80spam ◴[03 Sep 25 11:49 UTC] No.45114646[source]▶
Wow. I admit that I am not a native speaker, but this looks (or rather, sounds) VERY impressive and I could mistake it for hearing two people talking.
replies(2): >>45114849 #>>45114891 #
x187463 ◴[03 Sep 25 12:19 UTC] No.45114849[source]▶
The giveaway is they will never talk over each other. Only one speaker at a time, consistently.
replies(3): >>45114901 #>>45114974 #>>45115896 #
1. tracker1 ◴[03 Sep 25 12:26 UTC] No.45114901[source]▶
Fair enough... though it would be possible to generate that and edit to overlay the speech, introducing stuttering/pauses at the beginning and end of statements then edit the output to overlay the steps.

Would probably want to do similar to balance crossfade anyway... having each speaker's input offset from center instead of straight mono.