Prediction of hearing aid performance using the multiple model least squares technique

Measurement of noise and distortion in hearing aids is important for the design, fitting and assessment of these devices. In addition, it is imperative to test the hearing aids with speech signals to accurately predict their "real world" performance. In this paper, an adaptive system identification approach is taken to quantify the distortion and noise in a hearing aid. The hearing aid was modelled as a time varying autoregressive moving average (ARMA) system whose coefficients are estimated on a block-by-block basis using the multiple model least squares (MMLS) algorithm. Several speech-based distortion measures are derived from the modelling procedure which is shown to perform well in predicting perceptual judgements of hearing aid quality.