Dual Microphone Noise Suppression (Spectral Subtraction)

Hello Forum

I want to try and setup a dual microphone system that does some sort of subtraction in order to enhance speech quality. The aim is to implement a communication system on a wearable sportsvest - all school related. I am working with two omnidirectional mems-microphones, one attached to the chest area, and one attached to the back of the body. My thought process was that i could use the backmounted mic, to capture and remove any background noise from the final audio output, by subtracting it's audiooutput from the speech-microphone coming from the chestmounted mic. I've read a bunch of articles online about this principle being applied in Smartphone devices when used in hands-free mode.

I have tried the build-in beamforming algorithm as well as simple time-domain subtraction but none of these seem to be working very well.

I am testing the setup using a mannequin bust with a speaker mounted in the facial area playing speech recordings, while playing sinewaves from a smartphone placed 10 meters behind the bust. The actual sound quality is fairly decent, but the subtraction part doesn't seem to be working.

If somebody has any ideas on how to implement this sort of thing in Sigma Studio, you are very welcome to answer down below :-)

All the best,

Mathias

PS. I'm working with the AD1452 Evalutation board + PDM output microphones with sensitivity rating -26dB.

Attached some pictures of the current setup which is very much at the "proof-of-concept" stage. The speakers on the shoulders are for comms audio coming in and not the one referred to previously.

sigma studio

speech

noise_reduction

beamforming

dual-microphone