Hi,
I found english accoustic model very accurate & thought of extending it for the Hindi language.
I added new hindi words in the lexicon dictionary but the align command gave error that "There were phones in the dictionary that do not have acoustic models".
When I add sound files (very few) for the new words, accuracy of the final alignment is not good as new word acoustic data size is very small.
I feel that English model can recognize the syllables very accurately & might possible that these syllables can be used to recognize Hindi words.
Can it be possible to add new words without adding matching sound files?
Thanks,
Vikas Patel