Actual and expected transcriptions are cleaned to produce similar output. - Githubissues

Joll59 / d-ser-t

d-ser-t quantifies speech recognition accuracy of the MSFT speech service and/or user created MSFT custom speech service models.

2 stars 2 forks source link

Actual and expected transcriptions are cleaned to produce similar output. #42

Closed KatieProchilo closed 5 years ago

KatieProchilo commented 5 years ago

Both the expected and actual transcriptions are cleaned in the following way in cleanTranscription():

Lowercase
okay is replaced with ok

Only the expected transcription will throw an error in validateExpectedTranscription if it contains anything other than:

Lowercase letters
- Everything is cast to lowercase anyway, so humans working on transcription files don't need to worry about this.
Apostrophes
Spaces

Only the actual transcription is cleaned in the following way in cleanActualTranscription:

Commas, periods, and question marks are replaced with an empty string. (NEW)
" 's" is replaced with "s" (NEW)

HobbyProjects commented 5 years ago

I'll leave it up to you but we should consider adding tests. The technical debt will only continue to get bigger if we don't start from somewhere. :)