Analysis of compressive properties of the BioAid hearing aid algorithm

Abstract

Objective: This technical paper describes a biologically inspired hearing aid algorithm based on a computer model of the peripheral auditory system simulating basilar membrane compression, reflexive efferent feedback and its resulting properties. Design: Two evaluations were conducted on the core part of the algorithm, which is an instantaneous compression sandwiched between the attenuation and envelope extraction processes of a relatively slow feedback compressor. Study sample: The algorithm’s input/output (I/O) function was analysed for different stationary (ambient) sound levels, and the algorithm’s response to transient sinusoidal tone complexes was analysed and contrasted to that of a reference dynamic compressor. Results: The algorithm’s emergent properties are: (1) the I/O function adapts to the average sound level such that processing is linear for levels close to the ambient sound level and (2) onsets of transient signals are marked across time and frequency. Conclusion: Adaptive linearisation and onset marking, as inherent compressive features of the algorithm, provide potentially beneficial features to hearing-impaired listeners with a relatively simple circuit. The algorithm offers a new, biological perspective on hearing aid amplification.

Key Words:

• Hearing aid algorithm
• auditory model
• compression
• efferent reflex

Acknowledgements

The authors would like to thank Ray Meddis for continuous support and fruitful discussions and Dirk Oetting for providing the algorithm outputs of the reference multichannel dynamic range compressor. TJ was financed by DFG Cluster of Excellence EXC1077/1 “Hearing4all”.

Declaration of interest

No potential conflict of interest was reported by the authors.

Notes

Supplementary material available online.

Notes

1. http://www.bioaid.org.uk

2. A calibration of this system to set 0 dB FS to 100 dB SPL and applying a 40 dB insertion gain would be suitable, for, example, aiding a hearing impaired listener’s absolute threshold of 60 dB HL. Average sound levels in the legend of Figure 3 transfer then to 20, 40, 60, 80 dB and 100 dB SPL.

3. The parameter choices of the time-frequency analysis would, e.g. be suitable for aiding a hearing-impaired listener’s absolute threshold of 50 dB HL when calibrating the system such that 80 dB SPL input level corresponds to 0 dB FS. IC threshold is then at 60 dB SPL and DFAC threshold at −10 dB SPL. A 32 dB insertion gain would complete the fitting.

4. Distortion products in Figure 4(c) only seem to appear later the further away they are from the primary tones. Both the primary tones as well as the distortion products build up simultaneously shortly after the signal onset. However, since the distortion products are of less intensity, they cross the threshold of being visible in the grey scaling of the figure later than the primary tones.

5. The rMDC was set up to aid the same hearing loss (50 dB HL) as the BioAid algorithm in the time-frequency analysis. This means 32 dB maximum insertion gain were given for low input signal levels and insertion gain was decreased in the gain table to consist of 12 dB at 80 dB SPL input level.

6. The implementation is detailed in the source code on https://github.com/audioplastic/BioAid (available since 2012).