Speech Recognition Based on the Processing Solutions of Auditory Cortex

摘要

Speech recognition in the human brain depends on spectral analysis coupled with temporal integration of auditory information. In primates, these processes are mirrored as selective responsiveness of neurons to species-specific vocalizations. Here, we used computational modeling of cortical neural networks to investigate how they achieve selectivity to speech stimuli. Stimulus material comprised multiple pseudowords. We found that synaptic depression was crucial for the emergence of neurons sensitive to the temporal structure of the stimuli. Further, the subdivision of the network into several parallel processing streams was needed for stimulus selectivity to occur. In general, stimulus selectivity and temporal integration seems to be supported by networks with high values of small-world connectivity. The current results might serve as a preliminary pointer for developing speech recognition solutions based on the neuroanatomy and -physiology of auditory cortex.