XCore Exchange

Hi Everyone,

I'm recording voice signals using the Xmos (And Audacity on PC), 16KHz, and the SNR is quite low, around 10-12DB difference,

Distance of around 1 meter from the board and normal speaking loudness,

Had the same phenomena with some other board (Not Xmos and using CIC filtration),

(Quite disappointing since I would like to use the array's outputs with beamforming algorithms and they don't perform well with the current SNR),

Any idea if that is the case with MEMS technology?,

Thanks,

Hi,

This is strange, the SNR of the Akustica AKU441 microphones on this board are 63dB.
Have you checked your results against the Mic Array User's Guide : https://www.xmos.com/download/private/l ... rc1%29.pdf.
This gives a good overview of how the processing performs and gives some excellent graphs of the expected performance.

Are you sure you are not picking up noise from somewhere else ?

Best regards,
john

Hi John,
Yep, it is quite strange,
In order to make sure we are not picking up any noises we did more recordings today in an isolated acoustic recording studio,
Quiet segments: Average RMS power of -82.51 dB,
Speech segments: Average RMS power of -66.7 dB,
SNR: 82.51 - 66.7 => Around 15,
The speech was a in standard talk level 1 meter away from the board,
Not sure what might be wrong,
Is there any sensitivity parameter for the microphones that needs to be adjusted somewhere?
Any idea what we might be missing over here?
Thanks for your help!,
David

Hi David,
All you seem to be telling me there is the difference is levels between your speech and the background noise.
Have you measured the signal levels with a calibrated instrument and have you put in louder signals ?
Best regards,
John

Hi John,
Yes, we are trying to understand why is the difference small,
We haven't measured the signal levels - Did you mean using a sound level meter?,
And louder signals do improve the SNR - The louder and closer to the board we speak the better the SNR is,
We develop far field voice applications and would expect the board to record a relatively high SNR signals even in distances of around 5-6 meters without the need to shout,
At the moment the recorded ratio is around 1:6 (noise:speech levels) 1 meter away from the board,
Thanks,
David

Hi David,
Yes, you should use an SPL meter.
You seem to be measuring the signal to background noise level of the room.
This is not the same thing as the SNR of the microphone array.

> And louder signals do improve the SNR - The louder and closer to the board we speak the better the SNR is,
The SNR of the board doesn't change when you speak louder. What changes is the signal to background noise level of the room.

Best regards,
John

Hi John,
You are right - We measure the signal to background noise level of the room,
Using an SPL meter we get around 40db difference between signal (Standard speech 1 meter away) and background noise in a quiet recording studio,
And we were hoping to get the same difference when recording using the board but we get 15db,
(Noise amplitude levels of +/-10 and speech amplitude levels of +/-60)
Any idea how come that's what we get?, Is there any sensitivity parameter that we can control as in an analog microphone?,
Thanks,
Aviv