Résumé
The neural representations of acoustic features that differ in the location or timbre of the emitter elicit similar perceptions, suggesting the existence of a robust stimulus-response function between complex sounds and the activity of neural populations at all stages of the auditory system. This hypothesis is tested by decoding a random sound stream, using spike trains from a biophysical model of the auditory nerve and from large-scale recordings in the inferior colliculus, the auditory thalamus, and the auditory cortex of awake mice. At the level of individual neurons, the reliability of temporal and rate codes is found to decrease along the ascending auditory pathways. Rate coding is progressively favored with increasing independence of neuron frequency tuning. Firing in the auditory cortex is found to be synergistic, whereas that in subcortical areas is more redundant. Finally, combinatorial codes involving neural firing and neural silence within neuron pairs are shown to efficiently encode sound information, particularly in the auditory cortex. Overall, these findings reveal a progressive transformation of the neural code from an individual, redundant, and temporal code at the periphery to a more distributed rate-based code in the auditory cortex.