Localization Algorithms for Hearing Devices Influenced by Individual Variability in Ear Acoustics

Area of Science:

• Acoustics and Signal Processing
• Auditory Neuroscience
• Hearing Device Technology

Background:

• Head-related transfer functions (HRTFs) encode spatial audio cues vital for sound localization.
• Individualized HRTFs are crucial for accurate sound perception in hearing devices but are time-consuming to obtain.
• Substantial variability exists in head and ear acoustics, impacting HRTF performance across individuals.

Purpose of the Study:

• To develop and evaluate binaural and/or monaural algorithms for sound source localization using HRTFs in hearing devices.
• To assess the cross-subject variability and generalization performance of HRTF-based localization algorithms.
• To investigate the influence of gender on HRTF-based sound localization accuracy.

Main Methods:

• Constructed three algorithms (binaural/monaural) for hearing devices.
• Trained linear classifiers on HRTF databases (CIPIC, 3D3A) to predict sound locations.
• Evaluated cross-subject generalization and gender-specific performance.

Main Results:

• A "two-step" method achieved low horizontal localization error (1.0° CIPIC, 5.6° 3D3A).
• Vertical localization errors were higher (30.4° CIPIC, 36.5° 3D3A), improving when hemifields were separated.
• Classifiers trained on female HRTFs generalized better to males than vice-versa.

Conclusions:

• HRTFs contain essential localization cues for spatial hearing devices.
• Inter-subject variability, particularly gender, significantly influences the accuracy of HRTF-based localization.
• Developed algorithms offer insights for next-generation hearing aid spatial audio processing.