| ---
|
| license: apache-2.0
|
| pipeline_tag: audio-to-audio
|
| tags:
|
| - onnx
|
| - audio
|
| - speech-enhancement
|
| - noise-suppression
|
| - deepfilternet
|
| - deepfilternet3
|
| ---
|
|
|
| # deepfilternet3-onnx
|
|
|
| ## Overview
|
|
|
| This repository contains ONNX exports for DeepFilterNet3, a speech enhancement and noise suppression model.
|
|
|
| ## Source
|
|
|
| | Field | Value |
|
| |---|---|
|
| | Upstream model | [Rikorose/DeepFilterNet](https://github.com/Rikorose/DeepFilterNet) (DeepFilterNet3 weights) |
|
| | Upstream source revision | `dcbbe520263d1061693c4c4a56a6d6a917f30b25` |
|
| | Packaging source revision | `dcbbe520263d1061693c4c4a56a6d6a917f30b25` |
|
| | Export tool/script | `torch.onnx` export from DeepFilterNet3 checkpoint |
|
| | Quantization recipe | FP32 ONNX (`enc.onnx`, `erb_dec.onnx`, `df_dec.onnx`) |
|
| | Notes | Upstream weights are from the DeepFilterNet GitHub project; ONNX graphs are pinned at the packaging revision above. |
|
|
|
| ## Precision and Packaging
|
|
|
| Export tooling, precision, and quantization are recorded in the **Source** table above. This packaging mirror does not publish independent parity benchmarks; validate on your target execution provider before production use.
|
|
|
| ## Files
|
|
|
| | File | Description |
|
| |---|---|
|
| | `enc.onnx` | Encoder ONNX model |
|
| | `erb_dec.onnx` | ERB decoder ONNX model |
|
| | `df_dec.onnx` | Deep filtering decoder ONNX model |
|
|
|
| ## Intended Use
|
|
|
| Use these ONNX model files for speech enhancement and noise suppression experiments or integration work.
|
|
|
| ## Training Data
|
|
|
| Not documented in this repository. See the upstream DeepFilterNet project and paper for training details.
|
|
|
| ## Evaluation
|
|
|
| No repository-specific ONNX evaluation results are documented here.
|
|
|
| ## Limitations
|
|
|
| - DeepFilterNet usage examples commonly target 48 kHz audio; validate input requirements before deployment.
|
|
|
| ## License
|
|
|
| Apache-2.0. The upstream DeepFilterNet project is dual-licensed under MIT or Apache-2.0.
|
|
|
