Efficient auditory coding

TL;DR: It is shown that, for natural sounds, the complete acoustic waveform can be represented efficiently with a nonlinear model based on a population spike code, which shows striking similarities to time-domain cochlear filter estimates, have a frequency-bandwidth dependence similar to that of auditory nerve fibres, and yield significantly greater coding efficiency than conventional signal representations.

Abstract: Efficient coding theory posits that sensory systems are under strong evolutionary and developmental pressures to utilize highly efficient codes (Barlow, 1961; Atick, 1992; Simoncelli and Olshausen, 2001; Laughlin and Sejnowski, 2003). Using information theory, the basis of modern telecommunications, we have found that mammalian hearing follows this efficient coding principle. Neurons in the inner ear and the "spikes" with which they communicate form an efficient code for natural sounds in the environment (Smith and Lewicki, 2004a, 2005a, 2006). This shows for the first time that the theoretical principle of efficient coding can account for the detailed form of the auditory code, a significant milestone in developing a theoretical understanding of sensory coding. Additionally, the results of applying the same technique to speech coding suggest that the acoustics of speech are optimally adapted to this mammalian auditory code (Smith and Lewicki, 2004b, 2005b). Beyond these scientific issues, we show that a "spike"-like code may also lead to improvements to applications such as digital audio compression and telecommunications. In addition to our theoretical research, we sought to demonstrate efficient coding in human perception behaviorally. In a pair of experiments, we applied efficient coding theory to the problem of speech perception in individuals using cochlear implants (CI), for which there exist vast individual differences in spectral resolution and speech perception (Zeng et al., 2004b). We present a machine-learning method for CI filterbank design based on the efficient-coding hypothesis. Further, we describe a pair of experiments which evaluate this approach using noise-excited vocoder speech (Shannon et al., 1995). Participants' recognition of continuous speech and isolated syllables is significantly more accurate for speech filtered through the theoretically-motivated efficient-coding filterbank relative to the standard cochleotopic filterbank, particularly for speech transients. These findings offer insight in CI design and provide behavioral evidence for efficient coding in human perception.