Hi,
I have a lexicon where some words have multiple pronunciations. When I ran force alignment, Kaldi (gmm-align-compiled) will choose the best pronunciation given the utterance. How can I find the likelihood for each of the pronunciation variants?
Many thanks,
Vlad
--
Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/4d9a2752-9862-4440-a2f5-f01ee93de174n%40googlegroups.com.
Thanks for your answer Yenda!
Yes, I saw that I could get probabilities for each pronunciation in general using steps/get_prons.sh.
I would like to find the probability of each pronunciation for the given utterance.
gmm-align-compiled.cc script somehow picks the right pronunciation (in most cases) and as I understood it considers different pronunciations per word and picks the one which gives higher log-likelihood. So it seems that information I need should be somewhere but I can’t figure out how to access it.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/37618679-bd84-4635-8f0b-b20817a69016n%40googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/CAFReZQbS7s7zH6pYhBEG72PmH9ofgrMyet8pLu4TrgfaOCehAQ%40mail.gmail.com.