Abstract: Samples of an audio signal are converted into grouped frames of transformed transform coefficients, each frame having a plurality of regions (subbands). The power in each region is determined, quantized and encoded. A predetermined number of categorizations are applied to the transform coefficients in the plurality of regions. Each categorization assigns to each one of the regions a selected one of a plurality of different scalar nonuniform quantization step sizes. The step size is selected in accordance with the quantized power in the region and the quantized powers in all the regions. The quantized power encoded transform coefficients and an identification of the selected categorization are variable-length encoded and transmitted to a decoder. During decoding, the power of the transform coefficients in each of the regions is reconstructed from the quantized and encoded power. The selected one the categorizations used by the encoder is determined from the transmitted identification thereof.