Pre-trained PaSST-L with KD

v.0.0.10

add pre-trained model

Pre-trained models with KD

Added a pre-trained model, trained with KD using the setup in https://github.com/fschmid56/EfficientAT

0.0.8

v.0.0.8

add simple requirements file

Pre-trained PaSST-U and PaSST-B

Pre-trained PaSST-U and PaSST-B on Audioset

Pre-trained models on FSD50K and Openmic2008

• fsd50k-passt-s-n-f128-p16-s16-ap.642.pt pre-trained on FSD50K with structured patchout and no overlap map=0.642.
• fsd50k-passt-s-f128-p16-s10-ap.655.pt pre-trained on FSD50K with structured patchout map=0.655.
• openmic-passt-s-f128-10sec-p16-s10-ap.85.pt pre-trained on OpenMIC-2008 with structured patchout map=0.85.
• passt-s-f128-30sec-p16-s10-ap.473-swa.pt pre-trained on Audioset but supports inference up to 30-seconds.
• passt-s-f128-20sec-p16-s10-ap.474-swa.pt pre-trained on Audioset but supports inference up to 20-seconds.

Pre-trained models on ESC-50

• Pre-trained models on the 5 folds of ESC-50.
• The pre-processed ESC50 dataset for fine tuning.

v0.0.3-audioset

Pre-trained models with a smaller STFT hop

v0.0.2-audioset

Added more pretrained models

Pre-trained PaSST

v0.0.1-audioset

init