Patrick Li

Results 3 issues of Patrick Li

To make running in epoch mode easier, we redefine weights to effectively be 1. the number of times that dataset will be copied when weight > 1 2. fraction of...

Adding an extend audio task in ds_tool to create longer audio segments for eval

The whisper encoder has a max context of 30s of audio. This pr enables our model to support longer contexts by splitting long audio into chunks of 30sec (with the...