Validity of intelligibility measures for adaptive beamforming hearing aids

Abstract

Intelligibility‐weighted, averaged gain [P. M. Peterson, “Adaptive array processing for multiple microphone hearing aids,” MIT (1989)], GI, has been used as a figure of merit to evaluate adaptive spatial filtering algorithms for noise cancellation hearing aids. GI is computed from the input and output signal and noise power spectra. An estimate of GI, GI(w), can also be computed from the instantaneous directional frequency response of the adaptive spatial filter. GI(w) and GI are equivalent when the input signal and noise spectra are identical and the sound‐field nonreverberant; however, their relationship for other signal and noise spectra is unknown. Since GI(w) is computed by the filter coefficients, it can potentially be used as a criterion for design of spatial filtering algorithms that maximize intelligibility rather than signal/noise ratio. Studies were conducted to evaluate GI(w) as a predictor of intelligibility by comparing GI(w) with GI(w), AI, and actual measured intelligibility and sound quality. The two‐sensor Griffiths‐Jim beamformer with speech and noise inputs processed by a reverberant room simulation to produce the evaluation materials is used. The results of these evaluations will be presented, and the validity of these measures for use in the design of noise cancellation hearing aids will be discussed.