Intelligibility for Binaural Speech with Discarded Low-SNR Speech Components

Area of Science:

• Auditory Neuroscience
• Psychoacoustics
• Speech Perception

Background:

• Speech intelligibility in multitalker environments benefits from spatial separation of speakers.
• Binaural interaction, analogous to the Binaural Masking Level Difference (BMLD), may enhance target speech component detectability, especially in negative signal-to-noise ratio (SNR) conditions.
• This potential BMLD-like advantage could improve intelligibility by allowing detection of target speech components with disparate interaural cues.

Purpose of the Study:

• To investigate whether negative SNR target-speech components, potentially aided by binaural interaction, contribute to speech intelligibility in multitalker settings.
• To determine if a BMLD-like detection advantage plays a role in the intelligibility benefits observed with spatially separated speakers.
• To differentiate the contribution of negative SNR components from positive SNR components in spatial speech intelligibility.

Main Methods:

• A stimulus manipulation was employed where target speech components were selectively removed based on local SNR values.
• Criterion values for SNR were systematically varied to assess the impact on speech intelligibility for both spatially separated and collocated speakers.
• The study compared expected intelligibility degradation based on BMLD theory with actual observed degradation.

Main Results:

• Speech intelligibility degradation for spatially separated speakers was only observed at criterion values of 0 dB SNR and above.
• No significant degradation was found at negative SNR values, suggesting a negligible contribution of BMLD-like detection advantages.
• The results indicate that the spatial benefit for intelligibility is primarily linked to the separation of speech components at positive local SNRs.

Conclusions:

• The spatial benefit in multitalker speech intelligibility is not primarily driven by a Binaural Masking Level Difference (BMLD)-like improvement for speech components in negative SNRs.
• The observed spatial advantage is mainly attributed to the physical separation of speech components that have positive local SNRs.
• These findings suggest that auditory processing strategies in multitalker settings rely more on the segregation of clear speech elements than on exploiting subtle binaural cues in noisy conditions.