| --- |
| tags: [speech-enhancement, noise-cancellation, onnx, android, speech] |
| license: mit |
| --- |
| # DeepFilterNet3 — ONNX |
|
|
| Real-time noise cancellation for [speech-android](https://github.com/soniqo/speech-android). |
|
|
| - **2.1M params**, real-time at 48 kHz |
| - STFT: 960-point FFT, 480 hop |
|
|
| ## Files |
| | File | Description | |
| |------|-------------| |
| | deepfilter.onnx | Neural network (~8 MB) — pending export | |
| | deepfilter-auxiliary.bin | ERB filterbanks + Vorbis window (124 KB) | |
|
|
| ## Tensors |
| | I/O | Name | Shape | Type | |
| |-----|------|-------|------| |
| | in | feat_erb | [1, 1, T, 32] | float32 | |
| | in | feat_spec | [1, 2, T, 96] | float32 | |
| | out | erb_mask | [1, 1, T, 32] | float32 | |
| | out | df_coefs | [1, 5, T, 96, 2] | float32 | |
|
|
| --- |
|
|
| - **Guide**: [soniqo.audio/guides/denoise/android](https://soniqo.audio/guides/denoise/android) |
| - **Docs**: [soniqo.audio](https://soniqo.audio) |
| - **GitHub**: [soniqo/speech-swift](https://github.com/soniqo/speech-swift) |
|
|
