The get_motif_pairs method in datasets.py incurs IndexErrors sometimes. This is due to the "neighbor" motif pair not being found if the neighbor is in the last segment of audio in its respective audio file. The audio files are currently being truncated during the slice_seconds routine if the audio file does not evenly divide into 5 second chunks, which leads to some files missing their last chunk.
Ideally, we would pad the last segment so that it is retained and has the same length as the other segments. Perhaps we can have an option to either pad at the end, or "center" the last audio snippet with equal padding at both the beginning and end.
The
get_motif_pairsmethod indatasets.pyincursIndexErrors sometimes. This is due to the "neighbor" motif pair not being found if the neighbor is in the last segment of audio in its respective audio file. The audio files are currently being truncated during theslice_secondsroutine if the audio file does not evenly divide into 5 second chunks, which leads to some files missing their last chunk.Ideally, we would pad the last segment so that it is retained and has the same length as the other segments. Perhaps we can have an option to either pad at the end, or "center" the last audio snippet with equal padding at both the beginning and end.