If the user's audio results are saved in https://github.com/batumi/KartuliSpeechRecognition/issues/15 then if they edit the recognition results to what they actually said they will be able to train their own acoustic models (which will result in the most gain for accuracy).
Much better quality when run within the audio web service, where our ffmpeg command was more complete and it turned out to be more likely android wide band audio at 16k (which is wonderful):
If the user's audio results are saved in https://github.com/batumi/KartuliSpeechRecognition/issues/15 then if they edit the recognition results to what they actually said they will be able to train their own acoustic models (which will result in the most gain for accuracy).
Sample audio:
To do permit segment timed alignment in Praat, this is the command that needs to be added to the AudioWebService:
Much better quality when run within the audio web service, where our ffmpeg command was more complete and it turned out to be more likely android wide band audio at 16k (which is wonderful):