Binaural (pre)processing for contralateral sound field attenuation and improved speech-in-noise recognition

Understanding speech presented in competition with other sound sources can be challenging. Here, we reason that this task can be facilitated by improving the signal-to-noise ratio (SNR) in either of the two ears and that in free-field listening scenarios, this can be achieved by attenuating contralateral sounds. We present a binaural (pre)processing algorithm that improves the SNR in the ear ipsilateral to the target sound source by linear subtraction of the weighted contralateral stimulus. Although the weight is regarded as a free parameter, we justify setting it equal to the ratio of ipsilateral to contralateral head-related transfer functions averaged over an appropriate azimuth range. The algorithm is implemented in the frequency domain and evaluated technically and experimentally for normal-hearing listeners in simulated free-field conditions. Results show that (1) it can substantially improve the SNR (up to 20 dB) and the short-term intelligibility metric in the ear ipsilateral to the target source, particularly for speech-like maskers; (2) it can improve speech reception thresholds for sentences in competition with speech-shaped noise by up to 8.5 dB in bilateral listening and 10.0 dB in unilateral listening; (3) it hardly affects sound-source localization; and (4) the improvements, and the algorithms directivity pattern depend on the weights. The algorithm accounts qualitatively for binaural unmasking for speech in competition with multiple maskers and for multiple target-masker spatial arrangements, an unexpected property that can inspire binaural intelligibility models.