Detect sound of breaking glass

sadlah@yahoo.com.sg wrote:
> Hi guys. Thanks a lot for the excellent replies. This is actually a
> Senior Design class and we were assigned to do a project on DSP. Since
> we have National Instrument's Speedy-33 board, it would be easier to
> interface with it with Labview. What we're trying to build here is a
> home security system that monitors(through microphones) a glass panel
> to see if a break-in occurs.
>
> I guess what I wanted to know is if there is a specific pattern for the
> input that i will be getting off the microphone. because i think the
> generated waveform really depends on how hard i hit it or even the
> position on the glass where the initial hit occured.
>
> More specifically, how can I use Labview to achieve this? Or what type
> of signal processing would I have to undertake? I would say the
> timeline that I have is about 2 months.

Why not break some glass, record the sound, and analyse it?

Leon

"Leon" <leon_heller@hotmail.com> writes:

> 
> Why not break some glass, record the sound, and analyse it?
> 

Because he'll just be analyzing that particular window of that
particular glass at that particular time breaking. :-)

Ciao,

Peter K.

-- 
"And he sees the vision splendid 
of the sunlit plains extended
And at night the wondrous glory of the everlasting stars."

 

Peter K. wrote:
> "Leon" <leon_heller@hotmail.com> writes:
>
> >
> > Why not break some glass, record the sound, and analyse it?
> >
>
> Because he'll just be analyzing that particular window of that
> particular glass at that particular time breaking. :-)

So that he can get some idea of the characteristics of the sound. He
can then develop a suitable sytem, perhaps using neural nets, once he
has identified the relevant parameters. He'll need a lot of glass if he
does it that way. 8-)

Leon

"Leon" <leon_heller@hotmail.com> writes:

> So that he can get some idea of the characteristics of the sound. He
> can then develop a suitable sytem, perhaps using neural nets, once he
> has identified the relevant parameters. 

Sorry, Neutral Notworks don't work for me. :-)

> He'll need a lot of glass if he does it that way. 8-)

:-) Training over 100,000 epochs, yes, that is quite a bit!

Ciao,

Peter K.

-- 
"And he sees the vision splendid 
of the sunlit plains extended
And at night the wondrous glory of the everlasting stars."

 

Peter K. wrote:
> "Leon" <leon_heller@hotmail.com> writes:
> 
> 
>>Why not break some glass, record the sound, and analyse it?
>>
> 
> 
> Because he'll just be analyzing that particular window of that
> particular glass at that particular time breaking. :-)
> 

If the system really has to detect the acoustic sound, the neural net approach 
is pretty obvious (despite the info that someone has seen fit to patent it). For 
a training set, there are several online sources of sound effects, as a Google 
search will demonstrate. For example, this site:
http://www.sounddogs.com

has a whole category of glass sounds.

An alternative "advanced" strategy would be to identify the MPEG-7-style sound 
descriptors for breaking glass, and apply those to any detected sounds above the 
threshold level.

The sensitivity of the detector will have to be carefully set, as breaking-glass 
sounds may leigitmately crop up in the home not least via the TV, or even the CD 
player. There is a famous breaking-glass sample in "Babooshka" by Kate Bush (or 
maybe breaking plates, not sure!); it would be a pity if the alarm was triggered 
by a moment of musical nostalgia, or by Bruce Willis in full surround sound mayhem.

Richard Dobson