Ilan Malka

Speaker extraction using LCMV beamformer with DNN-based SPP and RTF identification scheme

The linearly constrained minimum variance (LCMV)-beamformer (BF) is a viable solution for desired source extraction from a mixture of speakers in a noisy environment. The performance in terms of speech distortion, interference cancellation and noise reduction depends on the estimation of a set of parameters. This paper presents a new mechanism to update the parameters of the LCMV-BF. A new speech presence probability (SPP)-based voice activity detector (VAD) controls the noise covariance matrix update, and a speaker position identifier (SPI) procedure controls the relative transfer functions (RTFs) update. A postfilter is then applied to the BF output to further attenuate the residual noise signal. A series of experiments using real-life recordings confirm the speech enhancement capabilities of the proposed algorithm.

Performance of fixed in-car microphone array beamformer under variations in car noise

Microphone array beamforming is being increasingly employed in the automotive industry for the suppression of car noise associated with tire friction, wind, and the vehicle engine. In many cases, this noise can be treated as stationary, facilitating fixed, non-adaptive statistically optimal beamformer implementation. This way, the noise properties can be inferred in highly controlled laboratory conditions, leading to a reliable, predictable, and simple beamforming solution. However, although the driving noise is stationary, its acoustic properties may depend on the type of the road on which the vehicle is traveling, e.g. concrete road versus asphalt or new road versus old. Hence, the performance of a fixed beamformer may suffer in a case where it is applied to a road type that differs from the one for which it was designed. In the current work, we study this effect for the widely used family of Minimum Variance Distortionless Response (MVDR) beamformers. The results suggest that the negative impact due to the noise mismatch may be significant in certain frequency bands. Nevertheless, a speech recognition experiment carried out on enhanced voice signals demonstrates that interchanging between the road types investigated here has, overall, a relatively mild effect on the recognition performance.