Speech enhancement using the microphone array and short-term spectral amplitude estimator

The paper presents a new adaptive beamformer employing updates for both the source and interference short term power spectral densities (PSDs). These PSD updates track the variations in the spectral content, thereby yielding a spectrally optimized constraint for each time instant. Consequently, probability density function models for the source, interference and noise covariance matrices are proposed for the estimation of the PSD parameters. An optimization problem is formulated based on the proposed model. Evaluation in a real car environment shows significant suppression levels for the noise and interference by employing the proposed beamformer whilst maintaining low source signal distortion.