My idea is to find a word in wav/mp3 files. It is just one word. I have a database with a few hundred songs.

I am looking for pattern matching (this is not an exact pattern) it can be an word i choose but only one.

  1. Can you recommend a pre-processing algorithm. you run the algorithm on the files?
  2. Can you recommend a real-time algorithm?

I have found this online

  • 1
    $\begingroup$ If the word is pronounced with a different voice than your template and/or if there is background music it'll be out of reach with the simple technique you have linked to. $\endgroup$ Commented Jun 28, 2013 at 23:11
  • $\begingroup$ let us a assume i have 400 songs and i want to find out the word "House". how can i do that? $\endgroup$
    – Gilad
    Commented Jun 28, 2013 at 23:38
  • 3
    $\begingroup$ This is a hard problem. The first step is to isolate the vocals from the background music. Some pointers: dsp.stackexchange.com/questions/1499/…. The second step is to identify words - the problem is called "keyword spotting" in the speech recognition community. The fact that the word will be sung rather than spoken makes the problem harder. I doubt you'll get decent performances using even state of the art algorithms. $\endgroup$ Commented Jun 29, 2013 at 0:06

2 Answers 2


If you are really interested in voice recognition. Check out Hidden markovs model. It is used by several softwares to recognize your words


The problem statement has large amount of data for processing. In order to get the exact pattern matching, Artificial Neural Networks/Adaptive Neuro Fuzzy Inference System tool can be used for training this large amount of data and the network will optimize the result while testing.


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.