Combining auditory preprocessing and Bayesian estimation for robust formant tracking
IEEE Transactions on Audio, Speech, and Language Processing
Encoding bandpass signals using zero/level crossings: a model-based approach
IEEE Transactions on Audio, Speech, and Language Processing
A model for complex sequence learning and reproduction in neural populations
Journal of Computational Neuroscience
| Hi-index | 0.00 |
This paper considers a space of possible temporal codes, surveys neurophysiological and psychological evidence for their use in nervous systems, and presents examples of neural timing networks that operate in the time-domain. Sensory qualities can be encoded temporally by means of two broad strategies: stimulus-driven temporal correlations (phase-locking) and stimulus-triggering of endogenous temporal response patterns. Evidence for stimulus-related spike timing patterns exists in nearly every sensory modality, and such information can be potentially utilized for representation of stimulus qualities, localization of sources, and perceptual grouping. Multiple strategies for temporal (time, frequency, and code-division) multiplexing of information for transmission and grouping are outlined. Using delays and multiplications (coincidences), neural timing networks perform time-domain signal processing operations to compare, extract and separate temporal patterns. Separation of synthetic double vowels by a recurrent neural timing network is used to illustrate how coherences in temporal fine structure can be exploited to build up and separate periodic signals with different fundamentals. Timing nets constitute a time-domain scene analysis strategy based on temporal pattern invariance rather than feature-based labeling, segregation and binding of channels. Further potential implications of temporal codes and computations for new kinds of neural networks are explored.