The spatial benefit in speech intelligibility and the role of source segregation mediated by binaural cues

* Presenting author
Day / Time: 19.03.2018, 13:40-14:10
Room: MW 1801
Typ: Vorkolloquium
Abstract: Listening with two ears provides a benefit in understanding speech in the presence of competing sources all placed at different locations. Having two ears provides a benefit in terms of signal-to-noise ratio (SNR) at one of the ears. This better-ear effect can even occur for a frontal target speaker and two interfering speakers at the left and right side of the listener. Due to the spectro-temporal sparseness of speech, better-ear glimpses of the target speaker (spectro-temporal intervals with beneficial SNR) can occur alternatingly in both ears. Furthermore, binaural cues may provide cues for segregation, allowing the auditory system to determine whether particular speech components belong to the target speaker. Experiments will be discussed that show that a glimpse-based metric of better-ear listening accounts well for the variability that is seen in speech intelligibility across different target speech tokens, presented at the same overall SNR, but that the amount of glimpses needed depends strongly on the nature of the interfering sources. It will also be shown that binaural cues indeed support improved stream segregation and speech intelligibility but that only binaural cues within the most salient glimpses are important, which may be helpful in complex acoustic settings with unreliable binaural cues.