Incremental offline decoding/alignment API needed

ReadAlongs / SoundSwallower

An even smaller speech recognizer / force aligner

Other

32 stars 4 forks source link

Incremental offline decoding/alignment API needed #37

Open dhdaines opened 1 year ago

dhdaines commented 1 year ago

For force alignment we are not actually doing live recognition, but the API inherited from PocketSphinx still assumes this. We need a better one, which:

Assumes that the entire input is available (though perhaps not all in memory)
Does CMN on a sufficient quantity of input to get good results
Emits progress events (which can be wrapped in an Observable)
Breaks the computation into small enough pieces to be run asynchronously in the main thread

dhdaines commented 1 year ago

In fact this may need to potentially support quite large input files...