# Hidden Markov Model in Speech
Different word can have different audio signal.
HMM can be trained on :
- phoenemes
- syllables
- words
- group of words
Training get a bit difficult with continous speech If we used words, we would need a HMM for every word combination.
So might be better to use phoenemes. If we have 40 phoenemes, we might hava total 40^2 hmms.
