Call Today 716.688.4675

Fault detection in microphone arrays for speech enhancement

Microphone array beamforming for speech enhancement and automatic speech recognition is prevalent in many real life application. The performance of algorithms deployed for such applications requires that the microphones are not faulty. The requirement in distributed sensor networks for the sensors to pick near identical signals is easily attained in microphone array topologies since the microphones are within cm of each other. However, unlike wireless sensor networks where portions of the received signal can be known apriori and be used to form correlates, the desired signal may be unknown in acoustic networks. A more general assumption, the speech signal plus the ambient noise being correlated, has to be used. An alternate will be to use a sounding tone on system startup for diagnosis. This also has a drawback since it cannot be utilized after startup. For instance, for a device that will be powered for weeks, there is a need to do a periodic checkup on the health of the microphones. We consider the use of the received signals at microphones, whether speech frames or noise frames, by utilizing the pairwise correlates. Suppose the received signal at the microphone is given as:

y_i[n]= s[n-\tau_i] + \alpha[n-\epsilon_i] +\nu_i[n]

where s[n] is the speech signal and \alpha[n] ambient noise and \nu_i[n] is i.i.d zero mean Gaussian noise. \tau_i and \epsilon_i are the respective delays of the speech and the ambient noise like air conditioner or fan. It is obvious that even in the absence of speech, there will be correlated room ambience noise recorded at each microphone. A threshold and a moving average can be deployed to create a decision mask for faulty and non faulty microphones.\\A sample performance of this metric is shown in Figure 1 below:

 

Detection of faulty microphone

Figure 1: Detection of faulty microphone 7

Our approach can detect multiple faults since each correlate is over two sensors.

VOCAL Technologies offers custom designed solutions for beamforming with a robust voice activity detector, acoustic echo cancellation and noise suppression. Our custom implementations of such systems are meant to deliver optimum performance for your specific beamforming task. Contact us today to discuss your solution!

More Information

VOCAL Technologies, Ltd.
520 Lee Entrance, Suite 202
Amherst New York 14228
Phone: +1-716-688-4675
Fax: +1-716-639-0713
Email: sales@vocal.com