AES Europe 2026

Despite the growing number of hearing-impaired workers
wearing hearing-aids in occupational settings,
understanding speech in multi-talker situations remains
challenging. This difficulty is particularly pronounced in
open-plan offices, where simultaneous talkers; room
reverberation are prone to degrade speech intelligibility.
While spatial cues are essential for segregating target
speech from competing sources, hearing-aids signal
processing may alter binaural information that supports
spatial hearing.
Accurate evaluation of hearing-aids performance is
therefore crucial. Objective speech intelligibility metrics
offer an efficient alternative to time-consuming listening
tests; however, their validity in complex spatial scenarios
involving hearing-impaired listeners remains unclear.
Monaural metrics such as HASPI account for individual
hearing loss but neglect spatial information, whereas
binaural metrics such as MBSTOI incorporate spatial cues
but are primarily designed for normal-hearing listeners.
This study evaluates the ability of existing objective
metrics to predict speech intelligibility for hearing-aid
users in multi-talker spatial environments. Listening tests
are conducted on 20 hearing-impaired participants fitted
with binaural hearing-aids. Four types of multi-talker
auditory scenes representative of open-plan offices are
reproduced using a loudspeaker array. They involve a target
speech, combined with diffuse noise; a localized
competing speech source. Objective measurements are
performed using an acoustic mannequin fitted with the
participants’ hearing-aids. HASPI; MBSTOI values are
computed from the binaural signals recorded at the eardrums
; incorporating individual hearing losses.
Objective predictions are compared with subjective
intelligibility scores,; an ablation analysis is
conducted to distinguish the effects of hearing loss
modeling from those of binaural processing.