Noise-invariant Neurons in the Avian Auditory Cortex: Hearing the Song in Noise

Abstract

Given the extraordinary ability of humans and animals to recognize communication signals over a background of noise, describing noise invariant neural responses is critical not only to pinpoint the brain regions that are mediating our robust perceptions but also to understand the neural computations that are performing these tasks and the underlying circuitry. Although invariant neural responses, such as rotation-invariant face cells, are well described in the visual system, high-level auditory neurons that can represent the same behaviorally relevant signal in a range of listening conditions have yet to be discovered. Here we found neurons in a secondary area of the avian auditory cortex that exhibit noise-invariant responses in the sense that they responded with similar spike patterns to song stimuli presented in silence and over a background of naturalistic noise. By characterizing the neurons' tuning in terms of their responses to modulations in the temporal and spectral envelope of the sound, we then show that noise invariance is partly achieved by selectively responding to long sounds with sharp spectral structure. Finally, to demonstrate that such computations could explain noise invariance, we designed a biologically inspired noise-filtering algorithm that can be used to separate song or speech from noise. This novel noise-filtering method performs as well as other state-of-the-art de-noising algorithms and could be used in clinical or consumer oriented applications. Our biologically inspired model also shows how high-level noise-invariant responses could be created from neural responses typically found in primary auditory cortex.

Mechanisms of noise robust representation of speech in primary auditory cortex

“…Although the synaptic depression model alone can account partly for the suppression of additive noise, the combined depression/gain control model is necessary to replicate the neural data in more complex distortions such as reverberation. We found that although the interaction between neuronal tuning properties and the spectrotemporal profile of a distortion is an important factor in how the neuron's response changes for distorted signals (6), the presence of a distortion in spectrograms reconstructed using the static STRF model (SN) (Fig. 5B) that the observed enhancement of the stimuli in neural data cannot be explained exclusively by static, linear spectrotemporal filtering.…”

“…Despite substantial changes in neuronal activity at the single-cell level (4,6), stimulus information encoded across the neural population remained remarkably stable across distorted stimulus conditions. This robustness was demonstrated by quantitative comparison of original clean spectrograms to reconstructions based on the neural population response to the distorted stimuli.…”

“…Several theoretical and experimental studies have postulated that the distribution of linear spectrotemporal tuning of neurons found in the auditory pathway could support enhanced representation of temporal and spectral modulations matched to those prevalent in natural stimuli (6,7). Others have attributed this effect to nonlinear response properties of neurons (8) and adaptation with various timescales (9).…”

Neural processing of natural sounds

“…These methodological advances have allowed auditory neuroscientists to make significant progress in understanding the nature of the auditory computations that are found in the ascending processing stream of both birds [30,41,48,51,[58][59][60][61] and mammals [38,[62][63][64][65]. Selectivity for natural sounds is already present at the level of the inferior colliculus (IC) in the sense that IC STRFs show temporal spectral features that are found in behaviorally relevant sounds [38,41,[66][67][68].…”

“…Sound mixtures, such as those created by a chorus of insects or a crowd in a loud restaurant, have also their own statistical signature: specifically, the structure that is present in the modulation power spectrum of isolated vocalizations is washed out in sound mixtures whereas the long-time average sound spectrum of isolated sound signals and their mixtures remain similar. Given that the modulation power spectra of background sounds differs from that of foreground sounds, a modulation filter bank -a set of filters in the spectral and temporal amplitude modulations domaintuned to these differences could be used to separate signal from noise resulting from sound mixtures and such mechanism might be in place in secondary auditory cortical areas [30]. Because the sound spectrum of mixtures and signal are similar, this task would be impossible with a simple frequency filter bank -a set of filters in the sound frequency domain.…”