Neuromimetic sound representation for percept detection and manipulation (Q2502577)

Summary: The acoustic wave received at the ears is processed by the human auditory system to separate different sounds along the intensity, pitch, and timbre dimensions. Conventional Fourier-based signal processing, while endowed with fast algorithms, is unable to easily represent a signal along these attributes. In this paper, we discuss the creation of maximally separable sounds in auditory user interfaces and use a recently proposed cortical sound representation, which performs a biomimetic decomposition of an acoustic signal, to represent and manipulate sound for this purpose. We briefly overview algorithms for obtaining, manipulating, and inverting a cortical representation of a sound and describe algorithms for manipulating signal pitch and timbre separately. The algorithms are also used to create sound of an instrument between a ``guitar'' and a ``trumpet.'' Excellent sound quality can be achieved if processing time is not a concern, and intelligible signals can be reconstructed in reasonable processing time (about ten seconds of computational time for a one-second signal sampled at 8kHz). Work on bringing the algorithms into the real-time processing domain is ongoing.