Efficient Quantization Of Statistically Normalized Vectors Using Multi-Option Partial-Order Bit-Assignment Schemes

In this paper we focus on new options for the efficient quantization of statistically normalized target vectors at low bitrates. This problem is fundamental to many low-rate speech and audio coder designs. Here many such coders follow a general principle of taking a structured speech or audio signal, applying a process of redundancy removal and then quantizing each of the resulting statistically normalized targets to a relevant distortion level. We look at this latter problem when some of these targets are to be quantized at very low bitrates ( 1 bit/target-scalar). The approach we take is to efficiently communicate a target-adaptive pattern of unequal bit-assignments (noise allocations) across each target. This can increase performance over an approach that has a constant noise allocation even when target vectors consist of independent and identically distributed (i.i.d.) scalars. We extend these schemes to multi-option schemes allowing further options to adapt and improve performance