Abstract
The human cochlea includes about 3000 inner hair cells which filter sounds at frequencies between 20 Hz and 20 kHz. This massively parallel frequency analysis is reflected in models of auditory processing, which are often based on banks of filters. However, existing implementations do not exploit this parallelism. Here we propose algorithms to simulate these models by vectorizing computation over frequency channels, which are implemented in “Brian Hears,” a library for the spiking neural network simulator package “Brian.” This approach allows us to use high-level programming languages such as Python, because with vectorized operations, the computational cost of interpretation represents a small fraction of the total cost. This makes it possible to define and simulate complex models in a simple way, while all previous implementations were model-specific. In addition, we show that these algorithms can be naturally parallelized using graphics processing units, yielding substantial speed improvements. We demonstrate these algorithms with several state-of-the-art cochlear models, and show that they compare favorably with existing, less flexible, implementations.
Highlights
IntroductionModels of auditory processing are used in a variety of contexts: in psychophysical studies, to design experiments (Gnansia et al, 2009) and interpret behavioral results (Meddis and O’Mard, 2006; Jepsen et al, 2008; Xia et al, 2010), in computational neuroscience, to understand the auditory system with neural modeling (Fontaine and Peremans, 2009; Goodman and Brette, 2010; Xia et al, 2010), in engineering applications, as a front end to machine hearing algorithms (Lyon, 2002; for example speech recognition, Mesgarani et al, 2006; or sound localization, May et al, 2011).These models derive from physiological measurements in the basilar membrane (Recio et al, 1998) or in the auditory nerve (Carney et al, 1999), and/or from psychophysical measurements (e.g., detection of tones in noise maskers, Glasberg and Moore, 1990), and even though existing models share key ingredients, they differ in many details
Figure 3A. shows an example of a complex cochlear model, the dual resonance non-linear (DRNL) model (Lopez-Poveda and Meddis, 2001), which consists of the filtering of an input – stapes velocity – by a linear and a non-linear pathway
infinite impulse response (IIR) filtering is fast on graphics processing units (GPUs), there is a bottleneck involved in transferring data to and from the GPU
Summary
Models of auditory processing are used in a variety of contexts: in psychophysical studies, to design experiments (Gnansia et al, 2009) and interpret behavioral results (Meddis and O’Mard, 2006; Jepsen et al, 2008; Xia et al, 2010), in computational neuroscience, to understand the auditory system with neural modeling (Fontaine and Peremans, 2009; Goodman and Brette, 2010; Xia et al, 2010), in engineering applications, as a front end to machine hearing algorithms (Lyon, 2002; for example speech recognition, Mesgarani et al, 2006; or sound localization, May et al, 2011).These models derive from physiological measurements in the basilar membrane (Recio et al, 1998) or in the auditory nerve (Carney et al, 1999), and/or from psychophysical measurements (e.g., detection of tones in noise maskers, Glasberg and Moore, 1990), and even though existing models share key ingredients, they differ in many details. While in simple models, filtering is essentially linear (e.g., gammatones, Patterson, 1994; or gammachirps, Irino and Patterson, 1997), a few models include non-linearities and feedback loops, such as the dynamic compressive gammachirp (DCGC; Irino and Patterson, 2001) and the dual resonance non-linear (DRNL) filter (Lopez-Poveda and Meddis, 2001), which are meant to reproduce non-linear effects such as level dependent bandwidth or two-tone suppression To simulate these models, many implementations have been developed, on software (O’Mard and Meddis, 2010; Patterson et al, 1995; Slaney, 1998; Bleeck et al, 2004), DSP board (Namiki et al, 2001), FPGA (Mishra and Hubbard, 2002), or VLSI (Watts et al, 1992) chips. A second problem with current implementations of auditory models is that, due to memory constraints, they are often limited in the number of frequency channels they can work on simultaneously, typically tens or at most hundreds of channels, while there are about 3000 inner hair cells in a human cochlea
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
