Speech enhancement for multimicrophone binaural hearing aids aiming to preserve the spatial auditory scene

Abstract

Modern binaural hearing aids utilize multimicrophone speech enhancement algorithms to enhance signals in terms of signal-to-noise ratio, but they may distort the interaural cues that allow the user to localize sources, in particular, suppressed interfering sources or background noise. In this paper, we present a novel algorithm that enhances the target signal while aiming to maintain the correct spatial rendering of both the target signal as well as the background noise. We use a bimodal approach, where a signal-to-noise ratio (SNR) estimator controls a binary decision mask, switching between the output signals of a binaural minimum variance distortionless response (MVDR) beamformer and scaled reference microphone signals. We show that the proposed selective binaural beamformer (SBB) can enhance the target signal while maintaining the overall spatial rendering of the acoustic scene.

Binaural speech enhancement with spatial cue preservation utilising simultaneous masking

“…Similarly to the SBB method [10], the proposed method consists of two processing phases: a) the classification phase of TF tiles into target-dominant and noise-dominant, and b) the enhancement phase where the BMVDR is applied to the target dominant TF-tiles, while in the noise-dominant TF tiles a scaled (with 0 ≤ g ≤ 1) version of the noisy signal is used in both HAs. Let the left and right input narrowband SNRs (NBSNRs) be given by [15] …”

“…The improvements consider two aspects. First, unlike the original SBB [10] which uses only one input NBSNR in the classification stage, our implementation of SBB uses both η in L and η in R in order to guarantee target dominance in both ears. Secondly, in the original SBB method [10], the scaling parameter g was selected as…”

“…As in [10] we use an average g (computed across the noise-dominated DFT bins) for each time-frame for both the proposed and the SBB methods to mitigate coloration of the residual noise. Hence, g is time-varying but constant over frequency.…”

“…In this section, the proposed method is compared with the SBB method [10] for τ = −50 : 0.5 : 50 dB, and the BMVDR-N method [8], [14] with N = 0 : 0.1 : 1. The comparison is done in two different noisy acoustic scenarios.…”

“…The BMVDR-N method, initially proposed in [8] and further investigated in [14], combines the output of the BMVDR with a portion of the noisy unprocessed signal to preserve the binaural cues of the noise. A slightly different approach was presented in [10], referred to as the selective binaural beamformer (SBB). This method uses either the BMVDR output or a suppressed version of the unprocessed noisy acoustic scene, depending on whether the target or the noise is dominant in a time-frequency (TF) tile.…”