Get per word ascr from pocketsphinx
hypotesisForSpeechAtPath() currently gives us the hypothesis for the whole phrase. How would I get the individual score (ascr) for each recognized word?
Example output from running pocketsphinx_continuous with -backtrace flag set true:
word start end pprob ascr lscr lback
HI 0 25 1.000 -244 0 1
JOHN 26 69 1.000 -654 0 1
IS 70 88 1.000 -296 0 1
EVERYTHING 89 137 1.000 -487 0 1
ALRIGHT 138 196 1.000 -465 0 1
Hi,
At this point there is no API in TLSphinx for this. I think it should be easy to write. Right now I don't have time available but looking at the source code for Pocketsphinx I find the implementation for what you need and seems to be really easy. Take a look here
Sorry I can't help you more with this.
I will close this for inactivity.
Reopen this one for anyone looking something fun to do 😄