Detecting dog barks

A potential customer requested a feasibiltiy study for a product
requiring detecion of dog barks in a real-time recorderd digital audio
signal. Functional requirements describe:

"Barks should be detected with a accuracy of at least 80%, while other
(possibly loud) signals like traffic, playing children, etc should
generate false positives in no more then 10% of the cases."

Any tips on algorithms or literature to get me started ?

Thanks,

Ico

>A potential customer requested a feasibiltiy study for a product
>requiring detecion of dog barks in a real-time recorderd digital audio
>signal. Functional requirements describe:
>
>"Barks should be detected with a accuracy of at least 80%, while other
>(possibly loud) signals like traffic, playing children, etc should
>generate false positives in no more then 10% of the cases."
>
>Any tips on algorithms or literature to get me started ?
>
>Thanks,
>
>Ico
>
>
Dog barks differently depending on they are hungry or angry.
I love dogs.
 

On 8/5/2010 7:45 AM, Ico wrote:
> A potential customer requested a feasibiltiy study for a product
> requiring detecion of dog barks in a real-time recorderd digital audio
> signal. Functional requirements describe:
>
> "Barks should be detected with a accuracy of at least 80%, while other
> (possibly loud) signals like traffic, playing children, etc should
> generate false positives in no more then 10% of the cases."
>
> Any tips on algorithms or literature to get me started ?

Barks in general, or a particular dog? Even a relatively narrow range of 
breeds would help. Given the yip-yip of a Chihuahua, the baying of a 
beagle, or the booming bass of a Great Pyrenees, I don't see enough 
common underlying structure to distinguish a dog from an odd motorbike.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

On 8/5/2010 7:45 AM, Ico wrote:
> A potential customer requested a feasibiltiy study for a product
> requiring detecion of dog barks in a real-time recorderd digital audio
> signal. Functional requirements describe:
>
> "Barks should be detected with a accuracy of at least 80%, while other
> (possibly loud) signals like traffic, playing children, etc should
> generate false positives in no more then 10% of the cases."
>
> Any tips on algorithms or literature to get me started ?

Barks in general, or a particular dog? Even a relatively narrow range of 
breeds would help. Given the yip-yip of a Chihuahua, the baying of a 
beagle, or the booming bass of a Great Pyrenees, I don't see enough 
common underlying structure to distinguish a dog from an odd motorbike.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

On 05 Aug 2010 11:45:47 GMT, Ico <usenet@zeev.nl> wrote:

>A potential customer requested a feasibiltiy study for a product
>requiring detecion of dog barks in a real-time recorderd digital audio
>signal. Functional requirements describe:
>
>"Barks should be detected with a accuracy of at least 80%, while other
>(possibly loud) signals like traffic, playing children, etc should
>generate false positives in no more then 10% of the cases."
>
>Any tips on algorithms or literature to get me started ?
>
>Thanks,
>
>Ico

Hello Ico,
  Wow, that sounds like an interesting project.
Your problem, I believe, falls in the category 
of "signal recognition" (a topic of which I'm 
shamefully ignorant).  I wonder if some of the 
techniques used for human "voice recognition" 
might be applicable to your problem.  My guess 
is that your "barking detection" problem is NOT 
an easy problem to solve.

It would be interesting to look at the time-
domain, and frequency-domain, plots of an audio 
barking dog signal.

Concerning the barking of my neighbors' dogs, 
I'd be willing to work on this project for free  
if it included: (1) detection of dog barking, 
and (2) upon detection, application of a severe 
electric shock to the rectums of my human neighbors.

[-Rick-]

On Aug 5, 9:10=A0am, Rick Lyons <R.Lyons@_BOGUS_ieee.org> wrote:
> On 05 Aug 2010 11:45:47 GMT, Ico <use...@zeev.nl> wrote:
>
> >A potential customer requested a feasibiltiy study for a product
> >requiring detecion of dog barks in a real-time recorderd digital audio
> >signal. Functional requirements describe:
>
> >"Barks should be detected with a accuracy of at least 80%, while other
> >(possibly loud) signals like traffic, playing children, etc should
> >generate false positives in no more then 10% of the cases."
>
> >Any tips on algorithms or literature to get me started ?
>
> >Thanks,
>
> >Ico
>
> Hello Ico,
> =A0 Wow, that sounds like an interesting project.
> Your problem, I believe, falls in the category
> of "signal recognition" (a topic of which I'm
> shamefully ignorant). =A0I wonder if some of the
> techniques used for human "voice recognition"
> might be applicable to your problem. =A0My guess
> is that your "barking detection" problem is NOT
> an easy problem to solve.
>
> It would be interesting to look at the time-
> domain, and frequency-domain, plots of an audio
> barking dog signal.
>
> Concerning the barking of my neighbors' dogs,
> I'd be willing to work on this project for free =A0
> if it included: (1) detection of dog barking,
> and (2) upon detection, application of a severe
> electric shock to the rectums of my human neighbors.
>
> [-Rick-]


Ha-Ha-Ha-...

I guess your neighbors don't read comp.dsp

Neither do mine

But my neighbors' dogs rarely bark - they are quite happy and
friendly, sometimes too friendly..

The key is to keep your dog well-fed, warm and happy

On Aug 5, 7:45=A0am, Ico <use...@zeev.nl> wrote:
> A potential customer requested a feasibiltiy study for a product
> requiring detecion of dog barks in a real-time recorderd digital audio
> signal. Functional requirements describe:
>
> "Barks should be detected with a accuracy of at least 80%, while other
> (possibly loud) signals like traffic, playing children, etc should
> generate false positives in no more then 10% of the cases."
>
> Any tips on algorithms or literature to get me started ?
>
> Thanks,
>
> Ico

Simple, get your own dog and attach a sound detector with a high
detection threshold to your dog. When your dog hears the distant dog's
barks, your dog will act as a bark repeater/amplifier and set off your
detector which will be rarely falsed because of the high threshold of
detection.

Clay

> I wonder if some of the 
> techniques used for human "voice recognition" 
> might be applicable to your problem.  

   There was "word spotting" in the 70ies
while such research was funded by the DoD.
   Originally human listeners that
had only limited training in Russian
would listen to tapes of intercepted radio
communications and try to detect key words
in the conversation. Then the interesting
tapes would be handed over to people that
could do a proper translation.
   The idea was doing the first step by
computer cheaper.

   Dog barks ( burst of coloured noise ? )
are not similar to human speech. If
there was only be a single bark and
background noise i doubt a human listener
could always detect it. One notices its a dog
barking because the signal is repeating.

MfG  JRD

Jerry Avins <jya@ieee.org> wrote:
> On 8/5/2010 7:45 AM, Ico wrote:
>> A potential customer requested a feasibiltiy study for a product
>> requiring detecion of dog barks in a real-time recorderd digital audio
>> signal. Functional requirements describe:
>>
>> "Barks should be detected with a accuracy of at least 80%, while other
>> (possibly loud) signals like traffic, playing children, etc should
>> generate false positives in no more then 10% of the cases."
>>
>> Any tips on algorithms or literature to get me started ?
> 
> Barks in general, or a particular dog? Even a relatively narrow range of 
> breeds would help. Given the yip-yip of a Chihuahua, the baying of a 
> beagle, or the booming bass of a Great Pyrenees, I don't see enough 
> common underlying structure to distinguish a dog from an odd motorbike.

All very true, these are the first things that came to mind when I heard
of the project, not an easy task at all.

I did some quick tests with various barking samples I gathered, and I'm
not sure if this can be done at all, given the requirements. I found
there is one thing common to all barks though:

* relatively broadbanded, easily spanning two or three octaves

* bursty, i have found non longer than 350 msec

* Envelope seems quite consistent, attack adn delay seem to be equally
  steep.

* there seems to be the same 'melody' in the fundamental often, going
  up and down for a few notes during the bark

I guess my approach will be to find bursts of interesting length to
start with, these will pick the interesting sounds from background
noise. Then inspect the bursts for expected envelope, and check for
spectral properties.

Interesting project indeed, I might report back with some first results.

Thanks for the feedback

Wire a test dog to a few sensors and see when he responds
to the audio.

Steve