problem in sound detetction

[malik_123]

i am to detect the sound of human voice and vehicles
i am using microphone and after measuring the average sound , i am detecting the the presence of some human or vehicle when the peak value of the
sound crosses the threshold
but , in open air , because of speed of air , false alarm are generated

do any one have idea how to avoid it ,

what if we use the parabolic reflector ,
will it be possible that the use of reflector may decrease the effect of air as air will not directly reach to mic
any good idea?

 

[Audioguru]

A windscreen for a microphone is made from soft foam plastic or fluffy cloth material.
Look at them in Google Images.

 

[toctory]

A band-pass filter from 100Hz~600Hz would be useful to got the human voice.

For vehicle, a accelerometer sensor might be better.

 

[Audioguru]

A band-pass filter from 100Hz~600Hz would be useful to got the human voice.

Human speech is from 50Hz to 14kHz. If you filter out all the important high frequency consonants then it sounds like cows.
Even a muffled-sounding telephone has a frequency response to 3.4kHz.
FM radio and TV goes to 15kHz.
A CD goes to 22kHz.

 

[doraemon]

Hello!

Voice detection in noisy signals is not a trivial task, but there are some
simple algorithms achieving this task with various success rates.
In some of the existing methods, the background noise is first extracted.
You need a spectral representation of background noise. Then you take an
inverse transform of this background noise and subtract it from the original
signal.
Then there are strategies to find criteria related to the presence (or absence)
of voice. For instance:
- Zero crossing rate;
- Power inside of the voice formants' bands. Using these spectrum characteristices
with thresholds for each band can be used to detect voice.
I have done this, but it was years ago and I am sure that new techniques popped
up in the meantime.

Dora.

i am to detect the sound of human voice and vehicles
i am using microphone and after measuring the average sound , i am detecting the the presence of some human or vehicle when the peak value of the
sound crosses the threshold
but , in open air , because of speed of air , false alarm are generated

do any one have idea how to avoid it ,

what if we use the parabolic reflector ,
will it be possible that the use of reflector may decrease the effect of air as air will not directly reach to mic
any good idea?

 

[sclim9]

Hi Dora,

You really good in speech signal processing. I would to ask you regarding the previous msg you hv answer on 'problem in sound detection.

firstly, i would like to ask how you extract the background noise? how to calculate the spectral representation of the background noise?

secondly, recently i was try to find the matlab code for zero crossing rate, i found it but i would like to count the zcr in frame.

If you have any matlab code to share wit me. you are welcome. I would like to say thanks advance.

Please help me on this. Thanks

Regards,
sclim9

 

[doraemon]

Hello!

There are a few algorithms described on the net to detect voice. I guess that's
what you should do first: find one, implement it and verify for instance that
a led blinks when you talk and not when you grind coffee.

Now, when you successed detecting voice, you have to make assumptions,
for instance that the background noise is stationary. For instance what works
extremely well is the background noise of an engine. It's used for instance in
phonic systems for helicopters. The noise is even stronger than the signal
but it works well because the characteristics of the noise are extremely
stable.

What you have to do is to use your voice detector as follows:

- When there is no voice, then take the background noise, calculate its FFT and
store the FFT. Refresh this as long as you have no voice.
- When there is voice, calculate the FFT of the signal, subtract the FFT of the noise
(which should be almost the same as the current noise because the noise is
stationary), and then calculate the signal back by IFFT.

As far as I remember, you can drop about 10 ~ 12 dB of background noise with
this method. But it depends on the noise...

Dora.

EDIT: Just after posting I noticed that it was in "Analog circuit design". I'm not
sure it's the right place...

Hi Dora,

You really good in speech signal processing. I would to ask you regarding the previous msg you hv answer on 'problem in sound detection.

firstly, i would like to ask how you extract the background noise? how to calculate the spectral representation of the background noise?

secondly, recently i was try to find the matlab code for zero crossing rate, i found it but i would like to count the zcr in frame.

If you have any matlab code to share wit me. you are welcome. I would like to say thanks advance.

Please help me on this. Thanks

Regards,
sclim9