Microsoft · Voice

VibeVoice 1.5B

Commercial use: conditions applyMIT License

The only voice model here that has been run end-to-end, and the only one built for long output: 600 seconds, against 60 for the others. Microsoft removed the inference code from GitHub after finding misuse, and later restored the repo without the synthesis code. The weights stayed up. The licence is MIT, but the model card says the model is intended for research and not for commercial deployment. Those two statements conflict; know both before you rely on either.

microsoft/VibeVoice-1.5B on Hugging Face audio.cpp

The licence

Summarized from the linked source documents, last checked 25 August 2026. Every claim links to its source; anything unconfirmed is marked unconfirmed.

MIT License

Commercial use: conditions apply

The licence is MIT, which permits commercial use; this is confirmed from Hugging Face's metadata. The model card states the model is intended for research only and not for commercial deployment. That statement is non-binding and does not override the licence grant. The model is shown as conditional because the publisher has stated in writing what they want.

Conditions

MIT itself imposes only the notice requirement.
The model card's research-only framing is the publisher's stated intent, not a licence term. Weigh it as a signal about support and future availability; it is not a restriction.
The card prohibits impersonating a voice without explicit, recorded consent, disinformation, and real-time voice conversion.

What about the audio you generate?

Generated audio carries an audible AI disclaimer and an imperceptible watermark by design. There is no ownership claim over output. The watermark matters if the output is meant to pass as a recording.

Correction to the shipped catalog: The shipped catalog says "MIT (unverified)". MIT is now verified. It is shown as conditional, not as an unqualified commercial "yes", because of the model card's research-only language and Microsoft's withdrawal of the inference code. Neither is a licence term.

Based on the licence text as of the date below. Licences change; the links carry the authoritative version.

Specs

Transcribed from the shipped catalog, audiocpp_models_catalog.json — that file is authoritative if a number here disagrees with it.

Vendor: Microsoft
Catalog id: audiocpp-vibevoice
audio.cpp loader family: vibevoice
Can do: Text to speech
Weights repo: audio-cpp/audio.cpp-gguf
Weights file: VibeVoice-1.5B-GGUF/vibevoice-1.5b-q8_0.gguf
Account needed to download: No
Max clip length: 600s
Sample rate: 24.0kHz mono
Approx. VRAM: 3.0 GB
In the picker by default: No — opt in from Settings
Run end-to-end here: Yes

Getting it

There is no per-model install. One native binary runs all of these, so getting a model is a download of its weights and nothing else — no virtualenv, no torch pin, no CUDA wheel to match against your driver.

From the CLI

soundbox model install audiocpp
soundbox pull audiocpp-vibevoice

The binary is fetched from audio.cpp's GitHub releases the first time you ask for it and cached under your own data directory. Soundbox ships no native binaries and vendors none.
Weights land in <models dir>/audiocpp/audiocpp-vibevoice/ rather than in the shared Hugging Face cache, so a download stays scoped to the model you asked for.
This one is opt-in: switch it on in the Models page's “Enable more models” section before it appears in the picker.

Worth knowing

24kHz mono, the lowest sample rate in the catalog. Fine for speech, obvious if you mix it against 44.1kHz material.
Verified: someone here has run it end-to-end. That is a claim about this adapter working, not about quality.

Sources

Primary sources: the model's own repository, the project it came from, and the conversion audio.cpp actually downloads. Never invented — an entry with nothing findable says so instead.

Licence details on these pages are summaries of the linked source documents, last checked 25 August 2026. They are not legal advice, and Soundbox is not affiliated with any model's authors — weights download from Hugging Face under the user's own account and terms. Company and model names identify whose work it is, and belong to their owners. The linked licence text is authoritative; commercial use should be checked against it, per model, before anything ships.