2
$\begingroup$

I know basics about DSP, and now trying to complete a project on Speech recognition. So I read as many resources as I found, and got some ideas. They are like:

  • take a sample sound and break it into small frames

  • calculate MFCC for each frames( 13 of them)

  • use DTW to match the testing and template pattern.

    Now, here are the things I am having trouble with-

  • 13 MFCC's for n frames leaves me with 13*n MFCC's. How exactly do I use DTW here?

  • DTW gives me a distance between two patterns. How do I determine if they match or not?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

At the very beginning let me warn you that DTW approach is suitable only for spoken word recognition. Nonetheless it is interesting as a basic exercise.

I assume that you have a database of training files (templates) and you already extracted MFCC's for those. Each training file contains one utterance for a given word class, i.e. "Hello", "Bye". I also imagine that each class has more than one example (plausible).

Here is what you should do during the recognition stage:

  • Once you have the recording for the whole word, you calculate the MFCC's for each frame. You want the recording of the whole word, not just part of it.
  • Now you calculate the distance between your recording and each of the templates in your database. In case of DTW you will be calculating the cost between each of 13 dimensional frames (simple distance metric/norm, i.e. Euclidean, Manhattan, etc.). Once the DTW algorithm is finished, you will end up with the distance value (upper right matrix entry as can be seen below) between your test sample and each of the templates.
  • The last step is to make a decision: to which class your test sample actually belongs to? You can either do it by picking the class of template with the minimum DTW distance. But what is even better, you can use the k-Nearest Neighbours for that.

enter image description here

Improve this answer
$\endgroup$
8
  • $\begingroup$ "Once you have the recording for the whole word, you calculate the MFCC's for each frame. You want the recording of the whole word, not just part of it." If I calculate MFCC's for each frame, then I have more than one MFCC vector? How do I match them with one of my templates? @jojek $\endgroup$
    – Nafis
    Commented Mar 2, 2016 at 18:07
  • 1
    $\begingroup$ Please read the second point, as well as basics of DTW. You must calculate the DTW distance matrix between each and every frame! Cost function is a norm. $\endgroup$
    – jojeck ♦
    Commented Mar 2, 2016 at 18:40
  • $\begingroup$ master @jojek i havent understand the second point of the last step, can you providing me some references about the implementation of KNN algorithm. $\endgroup$
    – Hamza TST
    Commented Jun 4, 2016 at 18:59
  • 1
    $\begingroup$ @jotadepicas: Actually these do have n and m elements each. $\endgroup$
    – jojeck ♦
    Commented Mar 31, 2017 at 14:52
  • 1
    $\begingroup$ @jotadepicas: Actually there is loads of examples in google. For example this one. $\endgroup$
    – jojeck ♦
    Commented Mar 31, 2017 at 15:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.