Effect Of Broadband Gain Frequency Shaping On Horizontal Localization Performance In Individuals With Normal Hearing

Authors: Sathish Kumar1, Arivudai Nambi2

1Manipal Academy of Higher Education
2Kasturba Medical College, Manipal Academy of Higher Education

Background: Localizing the sound sources is essential for spatial awareness and detecting warning signs. Hearing impairment often diminishes this ability, and earlier studies indicated that even with bilateral Hearing Aids (HA), the localization abilities cannot be completely restored. This phenomenon is observed even in normal-hearing individuals, indicating that HA signal processing itself can affect localization performance. The current study investigated one of the overlooked HA–related factors, the effect of channel interaction between the adjacent bandpass filters in multi-channel HAs on sound localization abilities.

Method: The localization performance was investigated on six individuals with normal hearing using 19 loudspeakers spanning -90 to 90 azimuths. The study has six conditions, including two versions of unaided (UAI – raw HA mic recordings and UA – without HA), Aided (Unitron HA), and three versions of broadband filters (BBF) implemented using MATLAB: (SF0 – without BBF, SF300 – BBF cutoff at 300 Hz, and SF1000 – BBF with cutoff at 1000 Hz)

Results: Repeated-measures ANOVA and post-hoc analysis indicated increased Root Mean Square (RMS) error in aided condition compared to the UA. However, no significant main effects were observed for signed bias.

Conclusion: The absence of a significant difference between the unaided and BBF conditions indicates that our custom-written code effectively preserves spatial cues. However, identifying the reasons for the positive results over the aided condition is challenging due to insufficient information regarding the signal processing employed in commercial HAs. The observed positive results can be due to the use of FIR filters, which introduce linear delay across frequencies, thereby preserving ITD. Additionally, our signal processing was carried out at high resolution, such as implementing compression at 0.001 dB steps and sampling signals at 44.1 kHz, which likely contributed to the preservation of ITD and ILD.