Q3Varnam - 2018-05-20

I have a own language model - non english, trained with about 50 words. Most of the words are two syllable sounds.

Keyword spotting works perfectly in finding individual words. very fast and accurate, works without missing the word each time.

However if I add phrases with 3 words - having a total of 4 or 5 syllables all together in the phrase.

It doesn't recognise instantly, have to say it a few times.

Also Keyword phrase spotting works only with pre-trained phrases, For ex:

Keyword1 is recognised accurately every time individually.
Keyword 2 is recognised accurately every time individually.

Keyword1 and Keyword 2 don't occur continuously in any of the trained sentences.

If I have a KWS phrase as "Keyword1 Keyword2" - it doesn't get recognised at all.

Is it mandatory to have KWS phrases to be pre-trained in the acoustic model?

Or am I missing something during the training?

 

Last edit: Q3Varnam 2018-05-21