A Binaural Front End for Speech Intelligibility Models: Application to the Hearing-Aid Speech Perception Index...

Area of Science:

• Auditory Neuroscience
• Speech Perception
• Acoustics

Background:

• Binaural hearing enhances speech intelligibility when sound sources are spatially separated.
• Existing monaural speech intelligibility models fail to predict this spatial release from masking.
• This limitation necessitates development of models that incorporate binaural cues.

Purpose of the Study:

• To propose and evaluate a binaural front-end processing model.
• To integrate this binaural front-end with monaural intelligibility models.
• To predict speech intelligibility in complex acoustic environments with spatial separation.

Main Methods:

• A binaural front-end was developed to generate enhanced monaural signals from binaural inputs.
• Two versions (stationary and time-dependent) of the front-end were tested.
• The Hearing Aid Speech Perception Index (HASPI) was used as the monaural model.
• Model predictions were compared against intelligibility scores from three datasets with normal-hearing listeners.

Main Results:

• The stationary front-end accurately predicted intelligibility with spatially separated stationary noise.
• Monaural HASPI underestimated intelligibility in conditions with better signal-to-noise ratio.
• Front-end predictions showed reduced accuracy for non-stationary and diffuse noise conditions.
• The time-dependent front-end overestimated intelligibility at low signal-to-noise ratios.

Conclusions:

• The proposed binaural front-end, particularly the stationary version, can significantly improve speech intelligibility predictions in spatially separated noise conditions.
• Integrating binaural processing with monaural models is crucial for accurate prediction of spatial release from masking.
• Further refinement is needed for non-stationary and diffuse noise scenarios, and for low signal-to-noise ratios.