| 摘要 |
Transform coefficients of audio signal samples are grouped into subbands. A predetermined number of subband bit allocation patterns (encoding categorizations), specifying parameters including quantization step size, dead zone, maximum bits, vector dimension, and a variable length code for each vector, are available. Subband bits are assigned in accordance with the quantized power distribution over the subbands, and the bit allocation patterns selectively chosen in accordance with a fidelity criterion and a maximum bit rate. The quantized power, transform coefficients and the selected bit allocation pattern (encoding categorization) identity are transmitted to a decoder, where, using the known encoding categorization, the coded data are inverse-transformed and dequantized into a reconstructed audio signal.
|
