Looking 4 Noise Reduction Algorithms for Speech based on ARM platforms

[Manuel]

Dear All,

as per subject using just 1 mike.

Any hints are wellcome.

take care,
X!

[h0nk]

Dear Manuel,

with only one mic You are limited to:
- Use a mic with (super) cardioid beam.
- The classic noise gate, which mutes the channel when the signal is below a threshold level.
- A Voltage controlled filter, which filters out the high frequency components of the noise and is controlled by the signal level.
At higher levels of the signal the bandwidth of the filter increases.
Both work only at near distances speaker to mic.

For really good (and impressive) results, You should really use at least two mics.
Do a search for "Andrea Filter".
This may be already running in Your notebook.
With two (or more) mics You could also do things called "Beamforming".


Best Regards

[Manuel]

Dear Honk,

in my case the voice in acquired in front of the mike at a distance of 1..2 mt not more.

Then there are two distinct moment, one with acoustic noise absence (the best) but unfortunately after a voice recognition noise will start in a position quite close (think a white noise that can be stationary), the problem is the intensity of the white noise. So you move from 20..40bB to 50..70dB. My issue is in the 50..70dB range....i can not shout too much to gain the right SNR.

Further hints and considerations are always helpful....

I would use a DSP.

Andrea ... impressive....let's discuss this.

take care,
X!

[h0nk]

Dear Manuel,

Dear Honk,

in my case the voice in acquired in front of the mike at a distance of 1..2 mt not more.

Then there are two distinct moment, one with acoustic noise absence (the best) but unfortunately after a voice recognition noise will start in a position quite close (think a white noise that can be stationary), the problem is the intensity of the white noise. So you move from 20..40bB to 50..70dB. My issue is in the 50..70dB range....i can not shout too much to gain the right SNR.

Further hints and considerations are always helpful....

I would use a DSP.

Andrea ... impressive....let's discuss this.

take care,
X!

At first: A distance of 1 to 2 meters is not(!) near.
In addition to Your voice You may get thr reverb from the room.

Have You already tried to use a cardioid or super-cardioid mic?
This would suppress environmental noise in the first place.
You never could expect this cancellation afterwards.
An automatic level control may also help.

The Andrea filter needs two(!) mics to do the adaptive beamforming.
And these mics should also have a cardioid pattern with their
maximum in the direction of the speaker.

I have not tested voice recognition behind such a filter.
YMMV.

You should also measure the spectra of Your voice and the noise
to get insights in possible filtering.
If they are in the same range, then You are simple lost.

Visual Analyser or Wave Spectra will do the job.


Good luck, and
Best Regards