PaSST OpenMIC features (ONNX)

ONNX export of the PaSST-S OpenMIC checkpoint openmic-passt-s-f128-10sec-p16-s10-ap.85.pt from kkoutini/PaSST (release v0.0.5, Apache-2.0), with its 32 kHz mel frontend built in. The weights are unchanged; this is a format conversion.

  • Input samples32k: float32 [1, 320000], the first 10 s of mono 32 kHz audio, zero-padded.
  • Outputs: features float32 [1, 768] (frozen representation) and scores float32 [1, 20] (OpenMIC sigmoid scores).
  • model.onnx SHA-256: 79142146b9b56a2c3c4fb14f5585f28d66ecd756aaefad6f89ff8eb50b820cd9
  • Source checkpoint SHA-256: dc229428753176e8be0373d25887116fc15b490af86f671cecf9ed76a0f287da

Used by Document Graph Explorer for a short-clip bass-hit detector.

Citation: Khaled Koutini, Jan Schlüter, Hamid Eghbal-zadeh and Gerhard Widmer, "Efficient Training of Audio Transformers with Patchout", Interspeech 2022.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support