Smartglasses processing

Short demonstration

1 desired speaker (user) + 3 undesired speakers (fixed locations) ; SNRin = -5 dB .

Long demonstration

In the article, we described experiments in which different signals were combined and processed with a number of algorithms. The desired signal was broadcast from the Head and Torso Simulator (HATS) and received by the 8-channel glasses mounted array. Similarly, a number of undesired signals were broadcast and received by the array. The signals were combined to create various scenarios which were used to test and compare several algorithms.

Desired Signal

desired signal - the average of the two omnidirectional sensors

Noise Signal

Two noise scenarios were created:

(a) three stationary speech sources, a recording of the three stationary speech signals

(b) a moving speech source, a recording of a single undesired speech signal

Both are stereo recording with each channel corresponding to one of the omnidirectional sensors.


Stationary scenario:

Unporcessed Monopole Average
Proposed Algorithm (without post-processing)

 Moving interference scenario:

Unporcessed Monopole Avarage
Proposed Algorithm (without post-processing)

Post-processing results:

The following table demonstrates the effects of postprocessing. The results pertain to the scenario with three static interferers. Two sets of parameters are used for the post-prcessing stage. The first set (termed “post1”) is more conservative and the second (“post2”) is more aggressive.

Proposed Algorithm (without post-processing)


Dovid Y. Levin, Emanuël A.P. Habets, Sharon Gannot, Near-field signal acquisition for smartglasses using two acoustic vector-sensors, Speech Communication, Volume 83, October 2016, Pages 42-53.
[ArXiv document (open access)];     [Journal (possibly pay-walled)]