Youtube ASR Crawler

Usage

• Each line in drama_list.txt is of the format [url] [playlist_name]; wavfiles and subtitle file will be saved to [data_dir]/[playlist_name]/[playlist_index].{wav,vtt}.

  • bash crawl_playlist.sh [data_dir] playlists.txt

• process.py then process .vtt files and process it in to kaldi-style data format.

  • python process.py [data_dir] dir/to/{text,segments,wav.scp,utt2spk} [--merge-consecutive-segments]

• Timestamps of some of the closed captions are not very accurate, so you might want to --merge-consecutive-segments then run something like segment_long_utterances.sh to obtain more accurate timestamp of each segments.