XCore Exchange

I just got my vf-3100-c43 and loaded the binary sample code . All worked well according to the Quick user guide. I also realized the noise removal worked perfectly well when music was played from host computer to Xmos speaker. ( processed voice from channel 1 - audicity , 99% of the background music was removed, only my voice stayed )

I then ran the same program, but turned on some other speakers in the room then recorded my voice. The processed voice and noise reduction were not effective from the external speakers.

May I hear from the community how to address the issue when noise from multiple sources, like a cocktail party , restaurant , market place environment.

Sam

Hi Sammok,
In the first case, the echo canceller is removing the audio played out from the speaker.
In the second case the beamformer locks onto the loudest sound detected because this is how it decides which signal to beamform on.
In the case of the circular array, while it is not possible to steer the beam in a specific direction, you could use a keyword detector and then lock the beam onto the position of the received keyword.
The linear array allows the definition of a beam direction and beamwidth.

Best,
John

thanks, your reply is very helpful..

I am finding a way to use the analog audio jack as the mic output monitor. (Its binary sample code makes jack as host computer speaker. ) I need to create a loop back. Can anyone give me a sample program ? tks

Hi sammok,
You could do this in UserBufferManagement(), in BeClear_Int.xc.
Best regards,
John

sammok@beam2hear.com wrote:
I then ran the same program, but turned on some other speakers in the room then recorded my voice. The processed voice and noise reduction were not effective from the external speakers.

May I hear from the community how to address the issue when noise from multiple sources, like a cocktail party , restaurant , market place environment.

Sam

I'm right now trying to do the same test. I'm using linear vocal fusion speaker. I want to check how many decibels of noise background it is able to reduce. In order to check this, I have steered the parameter beamangle to the direction of the speaker and in other direction (separated 60 degrees) I have a speaker emiting background noise. I run the build "1i6o2_lin33" and analizing channel 0 (processed voice signal, optimised for the human ear) it should have less noise background than channels 1:4 (raw microphones data). Is this the best way to check the background noise reduction that vocal fusion speaker is able to perform??

Thanks in advance.

Antonio

Hi Antonio,

The noise suppression functionality gives up to 15dB stationary and diffuse noise suppression.
The beamforming will probably give less benefit but I don't have a specific figure for you.
Be careful with your choice of "noise". I found the algorithms performed better on real noise (the output from and Air Conditioning Unit) compared to noise created in a DAW.

Best regards,
John