A joint synchrony/mean-rate model of auditory speech processing

摘要: This paper describes a speech processing system that is based on properties of the human auditory system. A bank critical-band filters defines initial spectral analysis. Filter outputs are processed by model nonlinear transduction stage in cochlea, which accounts for such features as saturation, adaptation and forward masking. The parameters were adjusted to match existing experimental results physiology periphery. output this delivered two parallel channels, each produces representations appropriate distinct subtasks recognition One path yields an overall energy measure channel can be identified with average rate neural discharge. appear useful locating acoustic events assigning segments broad phonetic categories. In other path, extent dominance periodicities at channel’s center frequency captured synchrony measure, representation enhanced contrast, relative mean-rate spectrogram. show formant peaks during sonorant regions, smooth transitions over time, well preserving prominences high-frequency region fricatives stops.