0
$\begingroup$
  1. For N words, we have N HMMs. In each of the HMMs, the states are the GMMs. Each GMM models unique phoneme (with or without belonging to a particular context). It is these HMMs (words) that constitute nodes of an FSN (sentence). For each sentence, there is one FSN.

  2. For N words, we have N HMMs. In each of the HMMs, the states are still the HMMs, and they model phonemes. Just like in the case of words, each phoneme has its own HMM. In each of these phonemes modelling HMMs, the three states are GMMs which model the beginning, middle, and end of a phoneme. It is these HMMs (words) that constitute nodes of an FSN (sentence). For each sentence, there is one FSN.

Which one out of 1 and 2 is the correct implementation of GMM-HMM (in Kaldi)? Are both ways, correct in theory?

$\endgroup$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.