Re: [LAD] Detecting one sample within another

From: Luis Garrido <luisgarrido@email-addr-hidden>
Date: Sat Nov 03 2007 - 22:39:20 EET

> Given a sample clip, I would like to be able to find in another audio file a
> part that sounds similar to the clip. So for example, given a cymbal sample,

The biggest problem is defining what is 'similar' . Then how 'similar'
you want to get.

If you want to find an exact replica of the model sample within the
larger audio file a cross-correlation analysis is very simple and
might yield good enough results.

If you want a system with a better generalization ability, perhaps you
can extract certain parameters of the signal and feed them to some
classifying system. You could for instance use a set of model samples
to train a neural network and see how it reacts to the whole file.

There are libraries that could help you with both approaches.

But you need a good definition of the problem you want to solve and
bear in mind that what seems simple for your human brain might be
extremely complex to replicate artificially. It depends on what kind
of results are you demanding of your system. You need also a solid
background in digital signal processing. There is some general
literature on the subject available on the net.

There is a lot of literature on automated signal classification since
it is a rather useful feature to have (voice recognition is a prime
example): books, IEEE papers...

HTH,

Luis
_______________________________________________
Linux-audio-dev mailing list
Linux-audio-dev@email-addr-hidden
http://lists.linuxaudio.org/mailman/listinfo/linux-audio-dev
Received on Sun Nov 4 00:15:03 2007

This archive was generated by hypermail 2.1.8 : Sun Nov 04 2007 - 00:15:03 EET