iky1e/DeepFilterNet2-MLX

DeepFilterNet2 — MLX

MLX-compatible weights for DeepFilterNet2, a real-time speech enhancement model that suppresses background noise from audio.

This is a direct conversion of the original PyTorch weights to safetensors format for use with MLX on Apple Silicon.

Origin

• Original model: DeepFilterNet2 by Hendrik Schroeter
• Paper: DeepFilterNet2: Towards Real-Time Speech Enhancement on Embedded Devices for Full-Band Audio
• License: MIT (same as the original)
• Conversion: PyTorch -> safetensors via the included convert_deepfilternet.py script

No fine-tuning or quantization was applied. Weights are converted directly from the original checkpoint.

Files

File	Description
config.json	Model architecture configuration
model.safetensors	Pre-converted weights (~8.9 MB, float32)
convert_deepfilternet.py	Conversion script (PyTorch -> MLX safetensors)

Model Details

Parameter	Value
Sample rate	48 kHz
FFT size	960
Hop size	480
ERB bands	32
DF bins	96
DF order	5
Embedding hidden dim	256

Usage

Swift (mlx-audio-swift)

import MLXAudioSTS

let model = try await DeepFilterNetModel.fromPretrained("iky1e/DeepFilterNet2-MLX")
let enhanced = try model.enhance(audioArray)

Python (mlx-audio)

from mlx_audio.sts.models.deepfilternet import DeepFilterNetModel

model = DeepFilterNetModel.from_pretrained("iky1e/DeepFilterNet2-MLX")
enhanced = model.enhance("noisy.wav")

Converting from PyTorch

python convert_deepfilternet.py \
  --input /path/to/DeepFilterNet2 \
  --output ./DeepFilterNet2-MLX \
  --name DeepFilterNet2

Citation

@inproceedings{schroeter2022deepfilternet2,
  title = {{DeepFilterNet2}: Towards Real-Time Speech Enhancement on Embedded Devices for Full-Band Audio},
  author = {Schr{\"o}ter, Hendrik and Escalante-B., Alberto N. and Rosenkranz, Tobias and Maier, Andreas},
  booktitle={17th International Workshop on Acoustic Signal Enhancement (IWAENC 2022)},
  year = {2022},
}