Models of musical timbre using cortical spectro-temporal receptive fields and temporal codes

Abstract

It has been customary to propose several ‘‘descriptive’’ definitions to quantify timbre as one of the basic attributes of sound. Multidimensional scaling techniques have proposed several acoustic dimensions to characterize the timbre of subsets of sounds. Yet, in the general case, timbre is an intricate percept difficult to relate to concise acoustic cues. Timbre probably reflects the dynamic interactions between spectral and temporal characteristics of sound that play a crucial role in identifying natural sources and convey a wealth of information about the environment. Recent physiological investigations revealed possible neural mechanisms to extract and represent the spectral and temporal information in complex sounds. At the level of the primary auditory cortex, spectro-temporal receptive fields exhibit a multiscale representation that captures many aspects of dynamic sound spectra. In the current study, we explore the potential correspondence between these cortical properties and the identification of musical instruments. We investigate the use of such representation as a quantitative descriptor of musical timbre, as validated by behavioral data from human subjects. Additionally, we examine biologically plausible coding strategies in this multidimensional cortical mapping involving spike-timing information.