0
$\begingroup$

I'm doing a project on audio classification, mainly focused on Bird sounds.

I would like to know which is the best similarity measure that I can adopt to check test signal with training data.

I intent to do the project in matlab.

$\endgroup$

2 Answers 2

1
$\begingroup$

I would like to know which is the best similarity measure that I can adopt to check test signal with training data.

Welcome to engineering. There is no universally "best" approach, as it always depends on your specific models.

Mathematically, as soon as you can define what "similarity" should mean, you might be able to derive something.

Assuming that bird's auditory senses kind of work similar to human hearing, I'd try with the usual audio comparison models, probably a basic MFCC thing.

As a comment: A former advisor of me once said:

A week in the lab can easily save you up to two hours of reading literature.

Really, there's a whole lot of papers, textbook introductions and wikipedia articles on sound comparison. Start with anything, and read forward. That's going to be faster than asking for a single answer to your acute problem without getting the bigger picture of how things work.

$\endgroup$
3
  • 1
    $\begingroup$ You mean: a week is a birdy forest? $\endgroup$ Feb 2, 2016 at 20:39
  • $\begingroup$ Conversely, you can spend all your time trying to find the "best" algorithm instead of getting your hands dirty... $\endgroup$
    – Emre
    Feb 2, 2016 at 23:55
  • $\begingroup$ @Emre I personally find the best engineers AND researchers do both: First get a solid overview about state of the art, and then try a few approaches themselves, starting with the simplest, working their way up till they find something sufficiently functional. $\endgroup$ Feb 3, 2016 at 12:52
1
$\begingroup$

As @Marcus Müller wrote, the best similarity remains an open question depending on the context. For birds which imitate human speech, you can have a look at: The transformation of birds sounds into "speech". Other references are: A procedure for an automated measurement of song similarity or Parametric Representations of Bird Sounds for Automatic Species Recognition. A look at IEEE Xplore (as suggested by @Marcus Müller), Scopus, or Google Scholar with queries like [similarity AND ("bird sounds" OR "bird songs")] can provide results.

Even better, ask your favorite university librarian, she/he knows where to look at.

$\endgroup$
5
  • 1
    $\begingroup$ I'd also like to point out that for technically relevant things like these, IEEExplore often is a valuable source :) $\endgroup$ Feb 2, 2016 at 14:21
  • 1
    $\begingroup$ True, and the paper Parametric Representations of Bird Sounds for Automatic Species Recognition comes from this source. Caveat, it is a paywall, and Google Scholar often more directly offers preprints $\endgroup$ Feb 2, 2016 at 16:10
  • $\begingroup$ On the other hand, I assume OP is a student at a university, and many of those have general subscriptions for IEEExplore. $\endgroup$ Feb 2, 2016 at 19:29
  • 1
    $\begingroup$ You are right. I should not point to GS only (laziness). I also remember that librarians can do an excellent job, especially with yooung researchers $\endgroup$ Feb 2, 2016 at 20:37
  • $\begingroup$ A hooray to the librarians out there! Really, if you're at a university library, ask. Librarian (at least in Germany) is a hard-earned qualification with a formal syllabus, and these people tend to know a lot. $\endgroup$ Feb 2, 2016 at 23:03

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.