search

hack4impact-upenn / transcribe4all

Painless speech-to-text transcription
MIT License
42 stars 10 forks source link

Intelligently splits flac file such that no individual file exceeds 100mb #51

Closed yoninachmany closed 8 years ago

yoninachmany commented 8 years ago

r @sandlerben @roornelas

IBM only accepts files that are 100mb in size -To catch all the words, repeats a 5 s section. -SplitFlacFile converts file to Wav, gets number of chunks including redundant 5 seconds using a helper function, and extracts audio segments of a given length using a derivation (see comment) and another helper function in process of mock testing!

sandlerben commented 8 years ago

@yoninachmany look at the diff and explain how it makes any sense? It seems like this branch is making changes to config only (changes which are no longer necessary since I made a config package).

yoninachmany commented 8 years ago

@sandlerben not completely sure what happened, code was pushed in this commit (https://github.com/hack4impact/transcribe4all/pull/51/commits/75b02b2506fba48b60f643bb49f94cdbe8265fa0), perhaps an issue came up when transcription.go was renamed to 'utils.go'

Anyway, I copied code from that commit into utils.go

sandlerben commented 8 years ago

@yoninachmany okay here is what I want you to do because the diff is still not right.

  1. git checkout split-file
  2. git fetch && git reset --hard origin/master
  3. git cherry-pick ef60c51ffac32df28610de9ef57e2ae7d0a28e17
  4. git push -f origin HEAD
sandlerben commented 8 years ago

@yoninachmany looks pretty good. Have you tested this locally?

yoninachmany commented 8 years ago

@sandlerben re "make sure this function is called where it needs to be called": how will this section (https://github.com/hack4impact/transcribe4all/blob/master/transcription/utils.go#L110-L128) need to be adjusted for a slice of flacPaths rather than a single flacPath?

sandlerben commented 8 years ago

basically run TranscribeWithIBM in a loop for each path