Effects of binaural rendering personalisation and reverberation on speech-on-speech masking

This study investigates the effect of head-related transfer function (HRTF) personalisation on understanding binaurally rendered target speech masked by interfering speakers in reverberant conditions. During a listening test, participants had to identify a correct colour-number combination from a virtual talker rendered in front of them while ignoring two interfering talkers positioned either in front or at the back. The sound was rendered with either an individual HRTF or one of two non-individual ones. These were selected for each participant as the best or the worstmatching from the same HRTF dataset, based on predictions of a computational auditory model for sound localisation. Two types of reverb from measured spatial room impulse responses (SRIRs) were applied to the speech: realistic dichotic reverberation decoded from 4th-order Ambisonic SRIRs or diotic reverb based on the omnidirectional Ambisonic channel IR as a baseline. Preliminary results show that realistic dichotic reverb improves speech perception when interfering speech is co-located with the target. No significant differences were observed across HRTF conditions on a group level, but individual HRTF-related performance differences exist, requiring further intra-subject analyses and data collection to characterise the individual results. Link to paper: https://aes2.org/publications/elibrary-page/?id=22671