Isolated word recognition system

isolated word recognition

Hey everyone!
Im a student of Electronics and Communication and, as part of our final year project, i am designing a MATLAB based Isolated word recognition system devised for Computer control. The size of the word bank is to be designed for 20 odd words.
The basic approach that i would be following is:
1. Importing the speech(word)/ audio file
2. Noise removal and framing into 20ms frames
3.Extraction of features using-
a) mel frequency ceptsral coefficients and
b) ZCPA
4. These features are then going to be recognised using a Hidden Markov Model
5. Post the recognition of the word, a signal will be generated for controlling the computer application

My queries are as follows:
1. Is this the right method of approach? Can all of this be executed on MATLAB?
2. Is it possible to use the HMM toolbox in matlab for this purpose?
3. I have collected a reasonable amount of information on the related topics but still find the matter insufficient. Are there any suggestions on what authors/publications i should read?
4. The computer control is for basic operations like opening and closing of a window, enabling the start option etc. As this doesnt directly fit under my field of study i know very little about it. Can anyone shed any light over this subject?

I hope my questions arent that burdensome. My inquisitiveness just portrays the enthusiasm towards the project.
All suggestions and ideas are most welcome.
Thanks

Ritwik Pandav

isolated word

hi everyone!!
iv got a bit of a problem. i understand HMM's but i dont know how does the model of an hmm translate to speech. i.e. what are the hidden and observable parameters, how do you compute the values of pi,A,B for a given word, how do you exactly create the model- the details...i have literature like rabiners work but its insufficient. if somebody could suggest anything, please do.
i would love atleast some active participation!!
cheers

isolated word recognition program in matlab

Hi, welcome to the project.

I’ve also started ‘simple word recognition’ project. I don’t have sufficient background on this topic. I’m running like headless chicken at the movement.
Are you gonna use HMM? My supervisor suggested me that ANN is better than HMM for SWR. In addition I looked at one student’s report, who did SWR project some years back, where it was suggested that ANN is better network for SWR.
I am not sure about this but I know HMM is more popular in speech recognition field.

BTW, I am new to Matlab; if you have or find useful feature extraction codes, plz help me.

Good luck.

ritwik pandav blog

kamehameha said:

hi everyone!!
iv got a bit of a problem. i understand HMM's but i dont know how does the model of an hmm translate to speech. i.e. what are the hidden and observable parameters, how do you compute the values of pi,A,B for a given word, how do you exactly create the model- the details...i have literature like rabiners work but its insufficient. if somebody could suggest anything, please do.
i would love atleast some active participation!!
cheers

Click to expand...

Hi
For HMM, I think the article of Rabiner is useful (specially for discrete HMM).
Otherwise, you can read the book "Spoken Language Processing: A guide to theory, Algorithm and System development" or HTK book.
Good luck
Muoinhohn