Sound recognition

[Moh]

Hi Everybody!

I am new to NI and LabVIEW ... I've just passed the fundementals exam and proceeding in the DAQ course.

I have been trying to do a task using MATLAB, but thinking now to move to LabVIEW and use the DAQs I have.. Have no clear idea if it is doable through LabVIEW..I know it might be early to do such a job relatively to my experience but if I find it doable, I'll work on it

 here it is :

I need to record bird sound and find out what is the bird.

As a first step, I have a couple of birds sounds and want to compare them. I'll leave the recording and database things to later.

Where should I start? What topics should I search in? and a question ahead all : Is that doable by LabVIEW?

Appreciate your efforts and time guys. Regards.

Moh

[chilly_charly]

Of course it's doable with LabVIEW ! However, I don't think that this is something you will be able to do in a couple of weeks...



If there were only two or three different sounds, I believe the problem would not be that difficult. However, since you are talking about data base, I think your project is more ambitious...



First you should learn about multivariate analysis (chemometrics...) since the recognition process will involve statistical calculations. Then you'll have to get proper sound records, where the bird songs can be clearly separated from background noise, and process these sounds to express them has convenient numeric data. You will need a way to normalize the sounds, since the signal amplitude will differ. You'll have to work in the frequency domain, and that's where multivariate analysis will come in, allowing you to select a number of significant frequencies, able to help you discriminate the bird species.



Another approach could be to use an artificial neural network. There as been an interesting post on the subject in the last 3 months.




Clearly, the most difficult part will not be LabVIEW, although the multivariate thing will keep you busy for a while... I have got a PCA example somewhere if you need it, but that will probably be far from sufficient...





Good Luck !

[Moh]

Many thanks Chilly

Actually I need to hand a proposal in a month or a bit less.. I think I'll start working on the multivariate analysis ..since i have no idea what neural networks are about .

If you meant that the database would be the probelm , then no problem, since i dont need to implement it now ..

I thank you again for your reply, and sure i'll ask you more when things get clear.

[DFGray]

Depending on the birds you are trying to identify, you may be better off using wavelets instead of FFTs.  As an extreme example, the north american cardinal emits short chirps of under a second duration.  Other birds do have longer calls, so FFTs may be more successful with them.  I do want to second chilly charlie's advice on how long it will take to do this.  You will end up playing a lot with the algorithms.  Start small and add complexity as you need to.  Don't be afraid to use multiple types of analysis and decomposition (e.g. FFT, wavelet, JTF, etc.).  Good luck.  It "sounds" like a fun project.

[Moh]

Well, it's fun but we have no much time

anyway, I appreciate ur advice, I'll keep here posted with what I do

thanks

Moh

Message Edited by Moh on 03-08-2006 08:36 AM

[rhijas]

Hi Moh, I don't know how helpful it will be for a bird song recognition program, but I am currently building a Musical Instrument Recognition Project that is based around analysing the harmonics that make up the instrument waveform.

My website is here: http://home.pct.edu/~rhijas55/SeniorProject/index.htm
Discussion: http://forums.ni.com/ni/board/message?board.id=170&message.id=173896&view=by_date_ascending&page=1

If you go to the PowerPoint presentations in my "Documents" page, it will give you an idea of what I am doing.

You could probably do something that recognises the notes that a bird plays over time. Kind of like if you where writing music. That would probably be more useful than an FFT since birds don't usually sing a constant note like instruments play.

[Moh]

Many thanks buddy.

I will spend tonight at your website 🙂

I think should find another way than FFT.

I have many things in mind. I will fully explain them soon .

Thanks again,

Moh