An auditory model-based approach to dynamic compression
Steffen Kortlang, Stephan Ernst, Volker Hohmann, Stephan D. Ewert
Medizinische Physik & Cluster of Excellence Hearing4all Universität Oldenburg
Sensorineural hearing loss typically results in elevated thresholds and steepened loudness growth with level significantly conditioned by a loss or dysfunction of outer hair cells (OHC). In hearing aids, amplification and dynamic compression aim at restoring audibility and at widening the limited dynamic range. However, success usually shows large inter-individual variability when considering, e.g., speech perception. In particular, hearing impaired listeners generally still have considerable problems in complex acoustic communication situations. Physiologically motivated approaches to dynamic compression may help by directly replacing damaged or absent OHC function.
An auditory-model based, fast-acting dynamic compression algorithm (MDC2) is introduced which aims at restoring OHC functionality. In the compression algorithm instantaneous gains are derived from a comparison of normal-hearing and hearing-impaired basilar-membrane input-output function. Frequency analysis and resynthesis is realized using a gammatone filterbank implementation [Hohmann, Acta Acustica 88, 433 (2002)]. Off-frequency component suppression is achieved using an instantaneous frequency estimate. The algorithm is fitted by estimating low-level gain loss (OHC loss) from adaptive categorical loudness scaling data and audiometric thresholds based on Ewert and Grimm [in: proc. ISAAR (2012), 393] and Jürgens et al. [Hear. Res. 270, 177 (2011)]. In a revised version of the compressor, a new stage prior to resynthesis was introduced, that applies a gain correction based on the amount of filter widening expected in the individual impaired system.
In a technical evaluation, the suggested compression algorithm and its different stages were compared to a conventional multi-band dynamic compressor with well-established gain prescription (NAL-NL1). Although the more complex model-based instantaneous compression comes along with slightly higher distortion components, other measures reveal its potential to restore perception for a wide range of input signals related to, e.g., the resulting signal-bandwidth dependent gain.
The considered algorithms were implemented for real-time operation using the ”Master Hearing Aid” signal processing platform [Grimm et al., Acta Acustica 92 (2006)]. Aided speech reception thresholds were measured for 30 hearing impaired subjects with moderate sensorineural hearing loss in a variety of spatial conditions. Moreover, after testing the system in a real world environment, the participants were asked to fill in a questionnaire including, for example, listening effort. As far as completed, aided speech reception thresholds in stationary and fluctuating noise were comparable to conventional compression strategies.
This work was funded by the BMBF project ”Model-Based Hearing Aids” (13EZ1127D) and DFG FOR1732.