22:02
Microsoft Open Source Innovation Audio Model VibeVoice-1.5B
Jin10 data reported on August 26, early this morning, Microsoft Research open-sourced the innovative audio model VibeVoice-1.5B. VibeVoice-1.5B has achieved multiple significant technological breakthroughs in the field of voice: it can continuously synthesize 90 minutes of ultra-realistic voice in one go, while most previous models could only synthesize less than 60 minutes of voice and would face issues like tone drift and semantic disconnection after 30 minutes.