Labelling the diphones (not the features, just the phonemes)

[Giorgia M]

I am trying to understand how the actual diphones get labelled (not the features, just the phonemes part). I am thinking of it as a pipelined approach which 1) looks up the recorded sentence’s words in the lexicon to get the pronunciations 2) using EM identifies the phonemes’ edges 3) finds the right place in the middle of the phonemes to then select diphones.

I believe there will be other things which are taken care of, like identify unwanted silences in the middle of the sentence, or carefully select the burst bit of a plosive.

I know this is simplistic, but is this in the right direction?
Thanks

[Simon]

We’ll look at this in detail in the lecture.

[Author]

Posts