Based on an analysis of a model of signal processing by the auditory system, we propose a local time-domain phase-correlation approach to the detection and categorization of sonorant speech features. In this approach, pitch pulses and formant frequencies are marked by characteristic patterns of phase-correlation in the output of groups of frequency-selective filters that correspond to the temporal sequence of firings of groups of nerve fibers in the cochlea. Algorithms for the detection of speech features based on the auditory model appear to improve upon conventional (i.e. spectrographic) techniques in several respects: they are resistant to additive noise, relatively independent of signal amplitude and spectral shaping of the input and speech-specific.