Individualized and noise-adaptive enhancement of speech intelligibility

The intelligibility of speech played back via audio reproduction systems is often impaired in noisy backgrounds. Ideally, algorithms enhancing speech intelligibility should be adaptive to the type and temporal variations of the noise, and also account for differences in individual listening preferences. While noise-adaptive algorithms have been investigated in several studies, individual preferences have not yet been addressed in this context. The current study investigated the inter-individual variability of normal-hearing subjects' preferences with respect to intelligibility enhancement in noise for communication applications using the AdaptDRC algorithm [Schepker, Rennies & Doclo, Proc. of Interspeech, Lyon, France, Aug. 2013, pp. 3577-3581], which has been shown to be highly effective in various types of background noise. Originally, the algorithm uses estimations of the SII to control spectral shaping and compression characteristics of the speech signal. In this study subjects were asked to adjust the parameters themselves based on their personal preferences at different SNRs in three types of background noises. The data are discussed with respect to the relation between individual listening preferences, generic model- based parameters and the predictability of individually preferred parameter settings, which would allow a complete individualization of the algorithm.