Perceptual Bit-Patterns based on Partial-Order Allocation Schemes with Application to Subband Speech and Audio Coding

In this paper we look at the problem of designing vector quantizers (VQs) when considering distortion criteria other than mean square error (MSE). In particular we look at audio-based supra-threshold distortion measures that are difficult to directly consider with traditional VQ design methods. Considered is the use of partial-order bit-allocation schemes as a way to optimize VQs for such distortion criteria. The product-code structure of a partial-order VQ naturally matches the underlying perceptual goal of distributing noise unequally across a signal vector. It also reduces the design problem to that of an integer-based combinatorial search, greatly simplifying the design process. In doing so it is important to note the approach exploits dependencies between scalars, thus maintaining many inherent VQ benefits. We show that the approach can be used to optimize and improve the performance of a product-VQ design with respect to the investigated perceptual criterion.