5
$\begingroup$

I am looking for a simple-minded audio example to demonstrate the abilities of an automatic bird species classifier. What I have in mind is an audio scene which is organized like this:

  • Bird A for 10-15 seconds
  • background noise for 0-10 seconds
  • Bird B for 10-15 seconds

The background noise may contain human-made sounds, audible weather, insects, etc. but no bird sounds. If the classifier works well, it should predict A (and only A), then nothing, then B (and only B). I don't mind what species A and B are as long as they are bird vocalizations.

Note that i want the example to be "natural"; i.e., acquired with passive acoustic monitoring, without capture nor machine playback. I am not looking for an artificial example made with an audio editor.

After searching through data repositories Xeno-Canto and Macaulay Library, a scene like that is surprisingly difficult to find. Most recordings have either one species in them, or multiple overlapping species.

Do you have an example of this to share? Or even better, can you point me to an automatic procedure that would allow me to search this on the Internet?

This question is related to Data sources for species co-occurrences in soundscapes? except that I am searching for sequential polyphony, as opposed to co-occuring sources.

Many thanks!

$\endgroup$

1 Answer 1

1
$\begingroup$

You can robustly test an automated classifier using one or several of the publicly available annotated datasets of soundscape data. Testing your classifier on a full dataset of natural sounds will include a wide range of scenarios involving multiple bird species (and sometimes none) such as the scenario you describe. Here are a few public annotated datasets to check out:

https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecy.3329

https://www.beei.org/index.php/EEI/article/view/5243

https://www.sciencedirect.com/science/article/abs/pii/S1574954123000432

https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecy.4047

$\endgroup$
1
  • 1
    $\begingroup$ I think items 1 and 4 from this list may contain appropriate data, though 2 & 3 seem to have single-species recordings. (There may in fact be multiple sounds per audio clip, but it's relatively unlikely, due to the data curation process.) $\endgroup$
    – Dan Stowell
    Commented May 2, 2023 at 12:34

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.