Representation of Complex Spectra in Auditory Cortex
Abstract
Natural sounds are broadband and dynamic. To understand their encoding in primary auditory cortex (AI), we have characterized the responses of units in AI with elementary versions of such spectra moving ripples. Ripples are broadband sounds with a sinusoidal envelope along the log frequency axis, that move up or down with a constant velocity. Speech spectra can be decomposed into a superposition of ripples with different densities and velocities. If AI units are linear, then it is possible to predict how a unit responds to any broadband dynamic stimulus by first measuring its responses to all elementary ripples (i.e., measure the ripple transfer function), and then superposing the responses to these ripples, each according to its weight in the input. We have successfully demonstrated the linearity of AI units in the past using ripples either stationary or moving only downward in frequency. The data described in this poster will show that transfer functions are also separable for up-moving ripples, but that the two transfer functions may well be different. Hence AI units are not always fully separable, but only separable by quadrant. We shall discuss the implications of these results and show examples of predicted and measured responses to speech.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1997
- Accession Number
- ADA483455
Entities
People
- Didier A. Depireux
- Jonathan Simon
- Shihab A Shamma
Organizations
- University of Maryland