Prev: make sudoku matrix
Next: Operation on Sub vector
From: Md Sahidullah on 6 Jun 2010 14:17
Dear Jorge,

You are dividing every speech file into 5 frames! I would like to suggest you to read the book 'Fundamental of Speech Recognition' By Rabiner & Juang (at least first few chapters). For HMM follow HTK tutorial.

I believe this is not a issue of MATLAB. I am requesting to post these kinds of algorithm specific issue to specialized community like comp.speech etc. You will get some better and quick response.

Regards
Sahidullah

\\"jorge " <zjebanie12345(a)gmail.com> wrote in message <hrm82h$6av$1(a)fred.mathworks.com>...
> Hi, I'm working on a speech recognition project. I'm trying to use Hidden Markov Models as a classifier.
> I have few question about it.
> For example I have a training set of speech files(10 different utterances, each utterance is pronounced 10 times). For each file Im doing feature extraction: every file is divided on 5 frames (constant number of samples - length(frame)==const) and for each frame I've got 12 MFCC coefficients.
> How to build a HMM to recognize 10 diffrent words having training data?
> I think, I must have 10 different HM models (for each word).
> What about structure of models?
> I think, every model must have 5 states (becouse every file is dividing to 5 frames).
> In each state I have 12 MFCC coefficients as an obsevations.
> In theory I know, how recognition proces will be proceded. But I dont know how do this in MATLAB.
> Do I need vector quantization for my features? (How can I do it?)
> How to build HMM models structure? (every state should have a loop on themselves, and transition to next state, without returning to previous state - am I right?)
> If you have any advice I will be greatfull.
 | 
Pages: 1
Prev: make sudoku matrix
Next: Operation on Sub vector