search

alumae / kaldi-offline-transcriber

Offline transcription system for Estonian using Kaldi
Other
226 stars 57 forks source link

README: Speaker ID process clarification #18

Open lkraav opened 6 years ago

lkraav commented 6 years ago

Perhaps the README could clarify what the expected process output is when speaker ID feature is enabled? What is supposed to look different in the text output compared to disabling speaker ID. Is it possible to give speakers names via some transcription configuration file, or is that post-text-editing work?

alumae commented 6 years ago

Yes, this needs to be clarified in the README.

Just to let you know, it only changes the names of the speakers in the output trs files, and the recognized speakers is a closed set of Estonian public figures who occur often enough in Estonian broadcast news (see for example the names in http://bark.phon.ioc.ee/tsab/p/play?trans=8840). It's not possible to change it or retrain it any way (currently). So it probably only interest you if you process Estonian broadcast news.

lkraav commented 6 years ago

Hehe, yeah this was useful information. Well, I'm running it on my custom audio, let's see what it comes up with. A retraining process would definitely be useful.