Kaldi is supposed to handle alternate pronunciations in lexicon.
I believe there was a bug in speech_kaldi_export.py where only the first form was selected, ignoring alternate pronunciation. With current dict-fr.ipa, the first pronunciation is often not the most frequent.
Kaldi is supposed to handle alternate pronunciations in lexicon.
I believe there was a bug in speech_kaldi_export.py where only the first form was selected, ignoring alternate pronunciation. With current dict-fr.ipa, the first pronunciation is often not the most frequent.