Zum Inhalt springen
ModelAufgelistetExtern
Browse files

Audiosronnx Deepfilternet

DeepFilterNet3 speech denoiser — the three ONNX graphs its own Rust/tract runtime uses, mirrored for the audiosronnx deepfilternet engine (loaddenoise("deepfilternet")).

Quellehuggingface.co/TigreGotico/audiosronnx-deepfilternet/tree/9e9bbc9ca4a09fe53440e223abf06f879d91f876
ArtOtherVersionDf decCBewertung

Modellquelle

Art
Other
Dateigröße
1,86–3,19 MB
Version
vDf dec
Aktualisiert
12. Okt. 2026
Quelle
Hugging Face

Quellenbeschreibung

audiosronnx-deepfilternet

DeepFilterNet3 speech denoiser — the three ONNX graphs its own Rust/tract runtime uses, mirrored for the audiosronnx deepfilternet engine (load_denoise("deepfilternet")).

graphinputsoutputs
enc.onnxfeat_erb[1,1,T,32], feat_spec[1,2,T,96]e0..e3, emb, c0, lsnr
erb_dec.onnxemb, e3, e2, e1, e0m[1,1,T,32] ERB mask
df_dec.onnxemb, c0coefs[1,T,96,10] deep-filter taps (freq-major)

48 kHz. The STFT/ERB frontend and synthesis come from DeepFilterLib (libdf, Rust — no torch); the mask application and deep filtering run in numpy. Inference is onnxruntime only.

Upstream (unmodified): https://github.com/Rikorose/DeepFilterNet — models/DeepFilterNet3_onnx.tar.gz

Validated: real noisy speech SI-SDR 2.82 -> 11.97 dB (+9.15), corr(clean, denoised) 0.97. Note: denoising degrades ASR (WER 0.533 -> 0.583) — use for perceptual cleanup, not before a recognizer. See the engine's conversion notes.

Quellen

1 Quelle
Hugging FacePrimärVersion 9e9bbc9ca4a09fe53440e223abf06f879d91f876
huggingface.co/TigreGotico/audiosronnx-deepfilternet/tree/9e9bbc9ca4a09fe53440e223abf06f879d91f876

Verifiziert 12. Okt.

Download files

Modellartefakte

3 Artefakte

df_dec.onnx

onnx · 3,19 MB · SHA-256 23114ce3b0f6…278a · Hugging Face

Download file

enc.onnx

onnx · 1,86 MB · SHA-256 7c5399d3da8a…1916 · Hugging Face

Download file

erb_dec.onnx

onnx · 3,14 MB · SHA-256 ab669a1d10af…d895 · Hugging Face

Download file

Quellenauszüge

2 Auszüge
Readme2. Aug.huggingface.co/TigreGotico/audiosronnx-deepfilternet

--- license: mit tags: [audio-to-audio, speech-enhancement, denoising, deepfilternet, onnx] --- # audiosronnx-deepfilternet DeepFilterNet3 speech denoiser — the three ONNX graphs its own Rust/tract runtime uses, mirrored for the `audiosronnx` `deepfilternet` engine (`load_denoise("deepfilternet")`). | graph | inputs | outputs | |---|---|---| | `enc.onnx` | `feat_erb[1,1,T,32]`, `feat_spec[1,2,T,96]` | `e0..e3`, `emb`, `c0`, `lsnr` | | `erb_dec.onnx` | `emb`, `e3`, `e2`, `e1`, `e0` | `m[1,1,T,32]` ERB mask | | `df_dec.onnx` | `emb`, `c0` | `coefs[1,T,96,10]` deep-filter taps (freq-major) | 48 kHz. The STFT/ERB frontend and synthesis come from `DeepFilterLib` (libdf, Rust — no torch); the mask application and deep filtering run in numpy. Inference is onnxruntime only. Upstream (unmodified): https://github.com/Rikorose/DeepFilterNet — `models/DeepFilterNet3_onnx.tar.gz` Validated: real noisy speech SI-SDR 2.82 -> 11.97 dB (+9.15), corr(clean, denoised) 0.97. Note: denoising degrades ASR (WER 0.533 -> 0.583) — use for perceptual cleanup, not before a recognizer. See the engine's conversion notes.

Note2. Aug.huggingface.co/TigreGotico/audiosronnx-deepfilternet

Source context: 0 downloads · 0 likes · Pipeline audio-to-audio · Repo TigreGotico/audiosronnx-deepfilternet