I have an audio signal where most of the signal of interest is above 400 Hz but it is corrupted by heavy wind noise. What type of filter would you recommend to try to remove the wind noise considering I have audio samples of just wind but I don't have uncorrupted samples of the signal of interest. A simple high pass seems to work ok but my boss wants me to try implementing some type of adaptive or spectral subtraction type filter.
6
-
$\begingroup$ Might be easier to enhance the region of interest if it has a more structure $\endgroup$– user28715Aug 9, 2017 at 0:13
-
$\begingroup$ Unfortunately it doesn't have much structure. The audio signal of interest is of cars driving by. $\endgroup$– BootstrapAug 9, 2017 at 0:35
-
$\begingroup$ So you want to suppress whoosh to hear whoosh (and tire sounds) Trucks and tanks are easier than cars. Using a wind screen while recording would help. Beamforming would help. Most cars don't growl like a Mustang $\endgroup$– user28715Aug 9, 2017 at 0:45
-
$\begingroup$ I second beaforming. Otherwise you could maybe suppress the noise by capturing the spectral profile of the noise and substracting $\endgroup$– FlorentAug 9, 2017 at 1:17
-
2$\begingroup$ To catch wind profile you can place one mic away from traffic and treat it as noise source. Then apply standard adaptive noise cancellation technique. Or for each measurement you need to know the wind direction, form a beam in direction of wind and another towards traffic. Then use first beam as noise source. $\endgroup$– learnerAug 9, 2017 at 5:57
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
Laminar air flow doesn't produce much in the way of sound. Turbulence, which is usually excited by interaction that is, the disrupted flow by an interacting object. Some of what is heard, is radiation from monopoles (dipoles, and quadrupoles) within the flow and some are resonances related to the object doing the disruption. You can often look at a tree in high wind and visually correlate the intensity of sound with the intensity of motion of leaves and limbs. If you stand near a steel tower in high wind, your will hear it "sing"
Part of what you record is the broadband excited vibration of the microphone's full structure. This is why a windscreen on microphone works. The screen dampens the wind incident on the microphones structure and is a poor radiator. One model is a random stream of impulses exciting a linear filter. The mic structure gives color to the wind sound.
This is a situation where a little care in data collection saves a lot of trouble down the line.
I read someplace once "Never invent gloves to fix a leaky valve"
You might also want to try to whiten the wind noise and see what happens. It would be similar to inverse filtering on the structure resonance.
Another characteristic of the vehicle sounds are the doppler shifts at the time when the car is at closest distance to the microphone.