VibeVoice-TTS · 1.5B
Long-form multi-speaker synthesis: up to 90 minutes and 4 distinct speakers in a single generation.
The model family
Six open checkpoints, one design language. Tap a card for the full detail.
5 result(s)
VibeVoice-ASR · 7B
Long-form recognition: 60 minutes in one pass, with who / when / what — speaker, timestamp and content.
VibeVoice-Realtime · 0.5B
Lightweight streaming: ~300 ms first-audio latency, streaming text input, robust to ~10 minutes.
VibeVoice-ASR-Streaming · 1.5B / 7B
Transcribes who said what while the audio is still arriving, emitting a line roughly every 2.9 seconds.
VibeVoice-ASR-BitNet · CPU
The ASR model squeezed from 4.62 GB to 1.58 GB — faster than real time on a few CPU threads, no GPU.