Adjust add/remove silence from segments:
clean_and_segment_data_nnet3.sh
it calls segment_ctm_edits.py, that one has options. e.g. --max-edge-silence-length can be increased from 0.5
could maybe increase --max-edge-non-scored-length too. But this depends on the result of listening to the data. e.g. if it's cut off tighter than 0.5 sec, it might be a different issue.
When doing data prep for training we need to segment long audios, the parameters should be adjust if the segments are chopped. It means it is missing the begining or the ending sounds
Comments
Post a Comment