We are half-way through implementing these bits of the Kaldi tutorial:
1) Create utterance-by-utterance transcript. Currently this is sitting as a text file with one verse ID number with its corresponding words available to it. However, if CMU Wilderness doesn't cut on verses, then this will need to be re-implemented.
2) Segments now found in data/Segments/ESSWYI/ESSWYI.segments
a. This doesn't include leading pause space. We'll need to double check that we're not missing massive chunks of the WAV files. How long is the file? Do the segments add up to roughly that length?
b. Where are the test files broken down into these segments from CMU? It's not verses, so does this cut off on pause space? are there words split across files?
3) wav.scp - Need to make a simple text file with locations for each split .wav file and its utterance ID. This has to be completed after 1) and 2) above. Also, we'll need to make a script to split the actual wave files on those cut marks once we're convinced we're not splitting words. Should be simple enough.
4) utt2spk - Done
5) spk2utt - Needs to take the inverse of utt2spk. Shouldn't take long
ssethia2
commented
5 years ago
We are half-way through implementing these bits of the Kaldi tutorial:
1) Create utterance-by-utterance transcript. Currently this is sitting as a text file with one verse ID number with its corresponding words available to it. However, if CMU Wilderness doesn't cut on verses, then this will need to be re-implemented.
2) Segments now found in data/Segments/ESSWYI/ESSWYI.segments a. This doesn't include leading pause space. We'll need to double check that we're not missing massive chunks of the WAV files. How long is the file? Do the segments add up to roughly that length? b. Where are the test files broken down into these segments from CMU? It's not verses, so does this cut off on pause space? are there words split across files?
3) wav.scp - Need to make a simple text file with locations for each split .wav file and its utterance ID. This has to be completed after 1) and 2) above. Also, we'll need to make a script to split the actual wave files on those cut marks once we're convinced we're not splitting words. Should be simple enough.
4) utt2spk - Done
5) spk2utt - Needs to take the inverse of utt2spk. Shouldn't take long