Update model card with FP16 precision and latency report

README.md

@@ -1,88 +1,49 @@
 ---
-license: mit
+license: apache-2.0
 tags:
   - speech-enhancement
-  - noise-reduction
+  - denoising
   - coreml
-  - apple-neural-engine
+  - apple-silicon
   - deepfilternet
-language:
-  - en
-  - multilingual
 library_name: qwen3-asr-swift
-pipeline_tag: audio-to-audio
 ---
 
-# DeepFilterNet3 - Core ML
+# DeepFilterNet3 — Core ML (FP16)
 
-Speech enhancement (noise removal) model converted to Core ML for Apple Neural Engine inference.
+Real-time speech enhancement model for Apple Silicon. Removes background noise from speech audio.
 
-Based on [DeepFilterNet3](https://github.com/Rikorose/DeepFilterNet) (Interspeech 2023).
+- **2.1M params**, FP16, ~4.2 MB
+- Runs on **Neural Engine** via Core ML
+- 48kHz native, 10ms frames
 
-## Model Details
+## Latency (M2 Max)
 
-| Property | Value |
-|---|---|
-| Parameters | 2.1M |
-| Model size | 4.2 MB |
-| Sample rate | 48 kHz |
-| Latency | ~40ms (20ms frame + lookahead) |
-| PESQ (DNS4) | 3.17 |
-| Compute target | Apple Neural Engine |
-| Framework | Core ML (mlprogram) |
-| Min deployment | macOS 14+ / iOS 17+ |
+| Duration | Time | RTF |
+|----------|------|-----|
+| 5s | 0.65s | 0.13 |
+| 10s | 1.2s | 0.12 |
+| 20s | 4.8s | 0.24 |
 
-## Architecture
-
-Signal processing (STFT, ERB filterbank, deep filtering) runs on CPU via Accelerate/vDSP.
-Neural network inference runs on the Neural Engine via Core ML.
-
-- Encoder: 4x SepConv2d + SqueezedGRU (256-dim, 3 layers)
-- ERB Decoder: SqueezedGRU + skip convs + sigmoid mask (32 bands)
-- DF Decoder: SqueezedGRU + deep filter coefficients (96 bins x 5 taps)
-
-## Usage with qwen3-asr-swift
+## Usage
 
 ```swift
 import SpeechEnhancement
 
 let enhancer = try await SpeechEnhancer.fromPretrained()
-let cleanAudio = try enhancer.enhance(audio: noisyAudio, sampleRate: 48000)
+let clean = try enhancer.enhance(audio: noisyAudio, sampleRate: 48000)
 ```
 
-CLI:
-
 ```bash
-audio denoise input.wav --output clean.wav
+swift run audio denoise noisy.wav --output clean.wav
 ```
 
-## Performance
-
-| Metric | Value |
-|---|---|
-| RTF (M2 Max) | 0.34 (3x real-time) |
-| 20s audio | ~7s processing |
-
 ## Files
 
-- `DeepFilterNet3.mlpackage/` - Core ML model (Neural Engine)
-- `auxiliary.npz` - Signal processing data (ERB filterbank, Vorbis window, normalization states)
+- `DeepFilterNet3.mlpackage` — Core ML FP16 model (Neural Engine)
+- `auxiliary.npz` — ERB filterbank, Vorbis window, normalization states
 
-## Conversion
+## Reference
 
-Converted from PyTorch checkpoint using `scripts/convert_deepfilternet3.py` in [qwen3-asr-swift](https://github.com/AufKlworworworworworkwork/qwen3-asr-swift).
-
-## License
-
-MIT (following DeepFilterNet3 original license)
-
-## Citation
-
-```bibtex
-@inproceedings{schroeter2023deepfilternet3,
-  title={DeepFilterNet: Perceptually Motivated Real-Time Speech Enhancement},
-  author={Schroeter, Hendrik and Maier, Andreas and Escalante-B, Alberto N and Rosenkranz, Tobias},
-  booktitle={Interspeech},
-  year={2023}
-}
-```
+- [DeepFilterNet3](https://arxiv.org/abs/2305.08227) (Interspeech 2023)
+- Part of [qwen3-asr-swift](https://github.com/ivan-digital/qwen3-asr-swift)