schism-audio/efficient-at-mn10-mlx

EfficientAT mn10_as (AudioSet) — MLX

EfficientAT mn10_as (MobileNetV3, width 1.0, 4.88M params, AudioSet mAP 0.471, Schmid et al. 2022) converted to MLX for Apple Silicon. Source checkpoint: mn10_as_mAP_471.pt from the fschmid56/EfficientAT GitHub Releases (v0.0.1). The models were distilled from PaSST transformers on AudioSet (527 classes).

Runs with the open schism-mlx package:

from schism_mlx.classify.efficient_at import EfficientAT

model = EfficientAT.from_pretrained("schism-audio/efficient-at-mn10-mlx")
for label, score in model.classify("song.wav", top_k=5):
    print(f"{score:.3f}  {label}")

Handles arbitrary-length audio in a single pass — the network is fully convolutional up to the global average pool, so no windowing or truncation is needed.

Conversion & verification

• Weights are the original float32 checkpoint, tensor-renamed and with convs permuted to MLX's channels-last layout. No quantization, no finetuning.
• Verified end to end against the genuine reference implementation (fschmid56/EfficientAT, models/mn + AugmentMelSTFT) on synthetic signals and real drum recordings from schism-audio/e-gmd: max logit difference 3.1e-5 with identical features (9.5e-5 end to end through the numpy frontend), identical top-5 everywhere. ~3.3 ms per 10-second clip on an M5 Max (RTF 0.0003).
• The mel frontend (pre-emphasis, 32 kHz, n_fft 1024 / win 800 / hop 320, 128 Kaldi-style mel bins, 0–15000 Hz, log(mel + 1e-5), (x + 4.5) / 5 normalization) is reimplemented in pure numpy from the reference's own AugmentMelSTFT code. No torch/torchaudio needed at inference time.

Precision note: on M5-class GPUs, MLX silently runs float32 matmuls at TF32-class precision (ml-explore/mlx#3860). schism_mlx disables this on import; if you load these weights with your own MLX code, set MLX_ENABLE_TF32=0 to match the reference implementation.

License

MIT, inherited from the source repository (code and weights). Model by Florian Schmid, Khaled Koutini, and Gerhard Widmer (CP JKU Linz); MLX conversion by schism-audio.

Citation

@inproceedings{schmid2023efficient,
  title={Efficient Large-Scale Audio Tagging Via Transformer-to-CNN Knowledge Distillation},
  author={Schmid, Florian and Koutini, Khaled and Widmer, Gerhard},
  booktitle={ICASSP 2023},
  year={2023}
}