search

wangyu / rethink-audio-fsl

Who calls the shots? Rethinking Few-Shot Learning for Audio (WASPAA 2021)
MIT License
40 stars 6 forks source link

How to label the Mix CLIPS form FSD-MIX-SED? #15

Closed chester-w-xie closed 2 years ago

chester-w-xie commented 2 years ago

"To label a clip, we consider all sound events within the 1s window. If an event overlaps with the window for more than 0.5s or half of the event duration, we add the corresponding class into the clip label. We then consider the number of classes within a clip as the level of polyphony with the assumption that it is rare to have short non-overlapping events within a 1s window. "

Can you provide a demo? Thank you very much!

wangyu commented 2 years ago

Added a demo function that does the labeling: https://github.com/wangyu/rethink-audio-fsl/blob/6e9626efc0fddadfe2f032e18d1794066a08c8b1/data/utils.py#L8