A metric for evaluating speech intelligibility and quality in hearing aids

Abstract

Noise and distortion reduce speech intelligibility and quality in hearing aids, but there are no metrics that encompass both noise and distortion. This presentation introduces new intelligibility and sound-quality calculation procedures based on the Speech Intelligibility Index [ANSI S3.5-1997]. The SII involves measuring the signal-to-noise ratio (SNR) in separate frequency bands, modifying the estimated noise levels to include auditory masking, and computing a weighted sum across frequency of the modified SNR values. In the new procedure, the estimated SNR is replaced by a signal-to-distortion ratio (SDR) computed from the coherence between the input and output signals of the system under test. The SDR replaces the system noise estimate with the combined noise and nonlinear distortion. For the procedure, the signal is divided into three regions comprising the low-, mid-, and high-level signal segments. The SII is then calculated for each region using the corresponding coherence. The three coherence SII values are then combined to predict the intelligibility and the sound quality for the device under test. The coherence SII procedure is shown to be accurate for both normal-hearing and hearing-impaired listeners for additive noise, peak-clipping distortion, and center-clipping distortion. [Work supported by GN ReSound (JMK) and the Whitaker Foundation (KHA).]