Forums

HMM Training

Started by namrata_reddy84 February 23, 2007
Hey everyone,
I'm working on training of HMMs for speech recognition using TIMIT
database. Can someone tell me if there is a matlab code available for
it?
What should the topology of the HMM(HMm structure) be? Should it be a
left to right(Baki's model) and a fully connected model( transitional
probabilities between all the states?

I appreciate any advice or help. Thank you

Namrata
Peace on you all,

you can use HTK toolkit for training and testing HMM. You can easily run it from matlab.

about the topology you asked about, you can use the traditional topology which is left to right with a number of states proportional to the length of the word being trained.

Best Regards
Abd-Elaziz

namrata_reddy84 wrote:
Hey everyone,
I'm working on training of HMMs for speech recognition using TIMIT
database. Can someone tell me if there is a matlab code available for
it?
What should the topology of the HMM(HMm structure) be? Should it be a
left to right(Baki's model) and a fully connected model( transitional
probabilities between all the states?

I appreciate any advice or help. Thank you

Namrata
Matlab statistical toolbox has functions like
hmmtrain() %for training a given HMM model
hmmestimate() %for estimating the probability of state transitions.

with regards,
Sinith.

On 2/24/07, abdelaziz abdelmoniem wrote:
>
> Peace on you all,
>
> you can use HTK toolkit for training and testing HMM. You can easily run
> it from matlab.
>
> about the topology you asked about, you can use the traditional topology
> which is left to right with a number of states proportional to the length of
> the word being trained.
>
> Best Regards
> Abd-Elaziz
>
> namrata_reddy84 >
> wrote:
> Hey everyone,
> I'm working on training of HMMs for speech recognition using TIMIT
> database. Can someone tell me if there is a matlab code available for
> it?
> What should the topology of the HMM(HMm structure) be? Should it be a
> left to right(Baki's model) and a fully connected model( transitional
> probabilities between all the states?
>
> I appreciate any advice or help. Thank you
>
> Namrata
>

--
sathyam vada dharmam chara