---
license: cc-by-nc-4.0
library_name: vokra
tags:
  - vokra
  - gguf
---

# facebook-denoiser (Vokra GGUF)

Converted to the Vokra GGUF format for [Vokra](https://github.com/ayutaz/vokra), a zero-dependency speech-AI inference runtime.

**This is a conversion, not a new model.** The weights are the upstream ones; Vokra re-packages them so its runtime can memory-map them directly. Credit for the model belongs upstream — see *Source* below.

## Files

| File | Size | SHA-256 |
|---|---|---|
| `facebook-denoiser.gguf` | 72.0 MB | `c0b23707a2f255b5eb108c5b08b92f310fede6870106e799b195282d6a375e74` |

## Usage

```bash
# Download (any HTTP client works — the file is a plain GGUF)
curl -L -o facebook-denoiser.gguf \
  https://huggingface.co/vokra/facebook-denoiser/resolve/main/facebook-denoiser.gguf
```

```bash
vokra-cli run --model facebook-denoiser.gguf --input input.wav
```

## Provenance

| Field | Value |
|---|---|
| Architecture | `facebook_denoiser` |
| Tensors | 48 |
| Upstream source | github.com/facebookresearch/denoiser (Meta real-time speech-enhancement waveform U-Net + LSTM, Defossez et al. 2020, CC-BY-NC-4.0 — owner §3.1 sign-off required, publish requires --allow-noncommercial) |
| Upstream licence | `cc-by-nc-4.0` |
| Licence class | `non-commercial` |
| Registry model id | `facebook_denoiser` |
| Vokra GGUF schema | 1 |
| Converted by | vokra-core 0.1.0-alpha.0 |

Every row above is read out of this file's own `vokra.*` metadata, so the card cannot claim something the artifact does not carry.

## Licence

The weights are distributed under **cc-by-nc-4.0**, unchanged from upstream. Conversion does not alter the licence, and your obligations run to the upstream author.

### ⛔ Non-commercial — you may not use this weight commercially

The upstream weight is **cc-by-nc-4.0**. It is republished here so the model can be evaluated and used for research, and the licence is unchanged by conversion.

- **Do not use this in a commercial product or service.** That restriction is upstream's, not Vokra's, and Vokra cannot waive it.
- Vokra's engine is Apache-2.0 and imposes no such limit — the limit is on **this weight**. Other models in this organisation are permissively licensed; check each one's card.
- Vokra's runtime refuses to load a non-commercial weight unless an explicit research flag is set, so this restriction is enforced at load time rather than left to the reader.

## Verifying this file

```bash
shasum -a 256 facebook-denoiser.gguf
# expect: c0b23707a2f255b5eb108c5b08b92f310fede6870106e799b195282d6a375e74
```