Quantization of line spectral parameters using a trellis structure

In this article, a low bit-rate low-complexity block-based trellis quantization (BTQ) scheme is proposed for quantization of line spectral frequencies for speech coding applications. The branches in the trellis diagram correspond to the LSF difference code-words and the states correspond to quantized LSF parameters. An efficient algorithm for the index generation (finding the index of a path in the trellis) is introduced. The proposed BTQ achieves the transparent coding quality at 23 bits/frame (1150 b/sec.) and offers a gain of 3 b/frame (150 b/s) and significant reduction in complexity, compared to the IS-641 split-VQ. An interframe BTQ scheme is also presented to exploit the redundancies between the adjacent frames. The interframe scheme is found to achieve an additional 50 b/sec reduction of the bit-rate and is based on adaptive block-based trellis quantization of the prediction residues.