Optimization of head -related transfer functions using subjective criteria.

Abstract

For many applications and devices directly involving an individual's perception, it is often desirable to incorporate the user's subjective preferences in the system design. Examples of such applications include digital hearing aids, home theater, and speech enhancement for cellular telephony. Rather than attempting to determine the system parameters through an objective correlate of the user's preferences, this research investigates the direct incorporation of subjective preferences in an iterative design procedure. This design approach is denoted Active Sensory Tuning (AST). The primary application considered in this work is the subjective optimization of head-related transfer functions (HRTFs) for synthesis of spatialized audio over headphones. While there are strong analogies between adaptive signal processing using an objective cost function and system design using subjective criteria, the latter requires consideration of practical limitations imposed by AST. In particular, the selection of an appropriate low-order model for the design space, quantization of the design space to perceptual just-noticeable-differences (JNDs), selection of search algorithms that are compatible with qualitative subjective evaluations of prospective designs, and the development of objective observer models to evaluate the expected behavior of such algorithms is addressed. The AST design approach and results are presented for two applications. The pseudo-signed algorithm (PSA) was developed for the enhancement of sinusoidal signals in the presence of noise. The subjective optimization of HRTFs was conducted using a modified genetic algorithm (GA), with design performance evaluated by the rate of and the quality of the final design. Under appropriate constraints on the design space, good quality HRTF designs may be achieved in a reasonable number of iterations.