0
$\begingroup$

I need to recognize just one bird. I am trying to compare the syllables extracted from the sound samples. As a result of the spectrogram function, the power spectral density is taken from the sound. They are 2 dimensional values. I like to compare this test signal with a known signal which is another 2 dimensional value. Could you please suggest what would be the best function to use for comparison. Please keep in mind that the syllable of the test bird could be time shifted or they closely match the original bird. I am attaching the sample spectrogram and its syllable extracted here after filtering.. First figure is the spectrogram before filtering. Second figure is the result of attempting to extract just the syllable.

Regards

$\endgroup$
  • $\begingroup$ Welcome to DSP.SE! This question is a very broad one. It would help to refine it if you said what you've tried... or what you're thinking about trying. Sometimes, a first step for these sort of problems is template matching. $\endgroup$ – Peter K. Sep 7 '15 at 13:19
  • $\begingroup$ Thanks Peter. I have done the log frequency transformation by taking the weightage more to the low frequency values which is interested for me.. Template matching is close to what I need. Where can I see how the function makeBinTemplate is written ? Unfortunately its not in matlab or python :( $\endgroup$ – Maria Sep 8 '15 at 14:55
  • $\begingroup$ The R package that defines it can be downloaded here. $\endgroup$ – Peter K. Sep 8 '15 at 15:04
1
$\begingroup$

A quick google for "bird voice recognition" yielded

https://peerj.com/articles/488/

which has an abstract describing that the authors refined the results they got using the old-school Mel-frequency cepstral coefficient comparison, as it has been common in (human) speech recognition for decades. I could start explaining what cepstrals and Mel-frequencies are, but there's so much literature out there that you literally just have to click read the first thing that hits you when you google for that. It's a relatively well-researched topic.

$\endgroup$
  • $\begingroup$ I have seen this document and once I tried MFCC. But it is part of feature extraction. Now I am more concerned on how to classify it after extraction of a feature.. or different classification algorithms.. Thank you $\endgroup$ – Maria Sep 8 '15 at 15:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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