Loudness Models for rehabilitative Audiology

Three different 3–channel dynamic compression schemes (automatic volume control, syllabic compression and compression limiting) which have previously been tested in the laboratory by Appell et al. [1995] were investigated in everyday life with five hearing–impaired subjects using a prototype digital hearing–aid. A battery of tests was performed containing categorical loudness scaling with narrowand broadband stimuli, speech intelligibility in noise using the Göttingen Sentence Test and quality assessment by paired comparison tests as well as a questionaire and an informal interview. Due to the carefully selected control conditions (i.e., unaided situation at roughly the same perceived loudness as in the aided situations, algorithms fitted with the same fitting rationale, same frequency response for all algorithms for a medium input level with a speech–shaped spectra) the differences across algorithms were only very small. Hence, no overall ‘winner algorithm’ can be derived from the current data. However, it is found that subjects with a small residual dynamic range and high speech reception thresholds (SRT) showed best performance in quality and speech intelligibility with dynamic compression whereas no clear–cut preference is found in the other subjects. From the current study it can be suggested that for low input levels a slow acting compression with a high compression ratio (i.e., automatic volume control) should be used to provide audibility at this input level range, whereas syllabic compression (small compression ratio) or even linear amplification seems to be beneficial at medium to high input levels. In any case compression limiting should be provided to prevent from high level signal peaks. This chapter has been submitted by Jens-E. Appell, Volker Hohmann, Birgitta Gabriel and Birger Kollmeier for publication in Speech Communication, special issue on speech processing for hearing aids.