microsoft / automatic-speech-recognition updated 6 months ago

VibeVoice-ASR

Name: VibeVoice-ASR
Author: microsoft

VibeVoice-ASR is a unified speech-to-text model designed to handle 60-minute long-form audio in a single pass, generating structured transcriptions containing Who (Speaker), When (Timestamps), and What (Content), with support for Customized Hotwords and over 50 languages.

Params

8.7 B

Context

—

Downloads 30d

695 K

Likes

1,253

Commercial use: allowed mit Not gated SAFETENSORS 51 languages View on Hugging Face ↗

Download history

daily snapshots · 10 days

696 K679 K

Jul 28Jul 31Aug 3Aug 6

Can you run it?

Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.

File	Quant	Size	Est. VRAM	Verdict on RTX 4090 · 24 GB
model.safetensors	bf16	17.3 GB	20.9 GB	⚠️ Tight — reduce context

Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.

Specifications

Architecture: VibeVoiceForASRTraining
Parameters: 8.7 B
Tensor type: BF16
Licence: mit
First seen on the Hub: 2026-01-21
Training datasets: undisclosed
Added to our catalog: 2026-07-28

Compare with

speaker-diarization-3.1 8.9 M whisper-large-v3-turbo 8.7 M whisperkit-coreml 8.3 M whisper-large-v3 5.5 M speaker-diarization-community-1 5.3 M whisper-base 4.8 M

Appears in

Best open coding models Commercial-use text models Models that fit in 12 GB