Call Today 716.688.4675

MVDR based two microphone beamforming with given desired signal direction

Beamforming solutions for two microphone arrays are highly desirable in applications which require blind source separation with some apriori knowledge of the signal direction and the interferer direction. Such solutions are used in applications such as mobile phones, automobiles to pick driver speech whiles suppressing front passenger speech etc. Using two microphones limits the achievable gains from conventional approaches. Further, the chosen algorithm has to bear in mind the computational burden in real time operating systems. The use of minimum variance distortion-less responses are curtailed in most applications because of the computational burden incurred in inverting the noise co-variance matrix. We present an approach which is based on MVDR but pre-computes the portions of the weights which requires inversion leading very high signal to noise ratio (SNR) improvements at minimal computational burden.

Consider a two microphones array as shown in Figure 1:

Two microphone array

Figure 1 Two microphone array

Consider the two single microphone speech signal:

y_i(t) = s(t-\tau_i) + n(t), i \in \{1,2\}

The frequency domain representation becomes:

Y_i(w) = S(w)e^{-j w \tau_i} +N(w), N(w) \sim \mathbb{N} (0,\sigma_n^2)

Now, suppose the desired direction of signals is 0^{\circ} whilst the interferer is from 180^{\circ}, similar to the situation that pertains to picking driver speech and rejecting front passenger speech. An MDVR based filter can be derived such the optimal filter w_{opt} can be approximated as:


\hat{w}_{opt} = \frac{G(\Gamma(w)) }{1+\cos{(2w(\tau_2-\tau_1))}} [1+e^{-jw(\tau_2-\tau_1)}, 2 \cos{(w(\tau_2-\tau_1))}]^{T}

where G(\Gamma(w)) = 1 - \mathbb{R}_e \{ \Gamma(w) \} \cos{(w(\tau_2-\tau_1))} -\mathbb{I}_m \{ \Gamma(w) \} \sin{(w(\tau_2-\tau_1))}.


The performance of such a filter is illustrated in Figure 2 with an SNR improvement of 18dB. It can also be seen that the speech from the undesired direction is attenuated by more than 20dB. It should be noted that no other nonlinear attenuations are applied to the processed speech.

Figure 2

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