Saturday, 15 November 2008

Developing a new Bird Song identification device – Help needed

I was on the forum on BirdForum and came across this post:

I am a 3rd year university engineering student who is developing a bird-identification device/software with my groupmates. We did some research and found out that a lot of the bird-identification devices are quite expensive and not very portable. So we thought we might be able to develop and build a cheap and portable bird-identification device (by processing the bird-songs) for bird-watchers and hopefully bird-watching will become an easier and cheaper activity for the public.
But before we can start building it, we have to know the needs and expectations for the device from the users, so we created a survey:
I would really appreciate it if you can take a minutes and fill in this really short survey for me =]
Thank you for your time!

I guess a lot of us birders would be interested in such a device. Please do help out with the survey if you are interested.


  1. I am a researcher working on a very similar problem. It is extremely difficult. I wouldn't expect to much from these guys. If a group of third year engineering students cracks this one I'd be very impressed. There are over 650 regularly occurring species in the United States alone to contend with. It requires extremely detailed analysis to separate birds to species using automated methods. If they're asking how many species they should be able to identify, and what would be useful information and one of the answers is "color" they've clearly no idea what they're doing. I've got some questions I'd like to ask them. Do they really think that its a possibility for them to ID 6000 species? Do they realize how much variability and fine structure is involved in separating bird calls? How do they plan to contend with variation within species (which can be greater than variation across species!), subspecies variation, geographic population variation, aberrational calls, juvenile calls, flight calls, gender specific calls, chip notes. What about birds with high song plasticity like house finch, purple finch, Brown-headed Cowbird, warbling vireo, bewick's wren, song sparrow, common starling, Northern Mockingbird, Long-billed thrasher, Brown thrasher, gray catbird, Lyrebird for heaven's sake!!! Never mind that, where are they going to get recordings for 6000 species? Do high quality recordings even exist for all the vocalizations of 6000 species? Are they going to use discriminate function analyses? Artificial Neural Networks
    ? Fuzzy Logic? CEPSTRAL? How are they quantifying songs parameters? How are they measuring song parameters? Off of short-time fourier transforms? How are they doing automated measurements of fine structure? For 6000 species? What kind of sample size per species are they going to have? How do they quantify note sweetness? How are they measuring trill rates? How are they dealing with ambient noise? How do they quantify and compare frequency and amplitude trending? Harmonic power? Preferential high-frequency attenuation in forest canopy? Song bandwidth contraction due to transmission distance and habitat type? I could go on and on here. These young chaps are extremely ambitious and sound extremely clueless. Just because they learned how to compute a fast-fourier transform, and whip up a simple Artificial neural network in their courses does not make them even remotely capable of accomplishing their stated objective. And cheaply and easily no less, good luck! Clearly they have no appreciation for the complexity they are faced with.

  2. Hi, Zachary. I am one of the students involved in this project. I am very sorry to have got on your nerves.
    The problem is that none of us in the group knows anything about bird identification (as you can tell). And we were just given this project with an extremely limited time span. As we were so lost in where to start, we made up this survey in the hope to get some feedback from some experienced birders like yourself.

    I do realise the extreme complexity of the problem we are facing. I think we are aiming to be able to demostrate at the end, automatic identification between 10 species of birds targeting a very small region.

    We have a week from now to complete a report saying that it is possible to identify birds by their birdsongs despite of whether we actually think it is possible or not. Although we do not have to go into too much technical detail in the signal processing yet. Please HELP!

    If you have some ideas about how to modify our survey, please send me the details. You can contact me by email.

    Please help us with our project and give us some guidance on where to go. I apologize on the behalf of our group for our ignorance on this topic.


Please leave a comment on what you liked or disliked about this post. Any thoughts on technique or improvements are welcome.