VibeVoice-ASR

microsoft-vibevoice-asr

fields

nameREQ	VibeVoice-ASR	github.com	2026-05-05
vendorREQ	Microsoft	github.com	2026-05-05
released_at	2026-01-21	github.com	2026-05-05
modality	audio	github.com	2026-05-05
access	open-weights	github.com	2026-05-05
context_window	65536	github.com	2026-05-05
parameters	7B	github.com	2026-05-05
benchmark_score
benchmark_name
announcement_url	https://github.com/microsoft/VibeVoice	github.com	2026-05-05

history · 8 fields · 8 revisions

name1 revision

VibeVoice-ASR current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

vendor1 revision

Microsoft current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

released_at1 revision

2026-01-21 current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

modality1 revision

audio current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

access1 revision

open-weights current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

context_window1 revision

65536 current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

parameters1 revision

7B current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.

announcement_url1 revision

https://github.com/microsoft/VibeVoice current github.com · 2026-05-05

VibeVoice-ASR: a 7B-parameter speech recognition model that processes up to 60 minutes (64K-token context) of continuous audio in a single inference pass, released January 21 2026. Built on Qwen2.5 base with continuous speech tokenizers at 7.5 Hz frame rate. Integrated into Hugging Face Transformers on March 6, 2026.