Please advise me on the more reliable confidence measure in PocketSphinx.
i would like to know the format of the score returned by ps_get_hyp(ditcKwsBeObject->ps, &score, &uttid) . how can i convert this score to a probability measure?
I can also get a simple posterior probability for each word using ps_seg_prob(seg, &ascr, &lscr, &lback) and average the posterior probabilities, but is this a good confidence measure compared to the score returned by ps_get_hyp for the whole utterance?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
There is no choice as you might think. The posterior probability returned by ps_seg_prob or by ps_get_prob is the approximation of the confidence measure. The acoustic score returned by ps_get_hyp is a value without any physical meaning.
Please advise me on the more reliable confidence measure in PocketSphinx.
i would like to know the format of the score returned by ps_get_hyp(ditcKwsBeObject->ps, &score, &uttid) . how can i convert this score to a probability measure?
I can also get a simple posterior probability for each word using ps_seg_prob(seg, &ascr, &lscr, &lback) and average the posterior probabilities, but is this a good confidence measure compared to the score returned by ps_get_hyp for the whole utterance?
There is no choice as you might think. The posterior probability returned by ps_seg_prob or by ps_get_prob is the approximation of the confidence measure. The acoustic score returned by ps_get_hyp is a value without any physical meaning.
See on page 33:
http://www.mmk.ei.tum.de/publ/pdf/diss/fabian.pdf
or any ASR textbook.