An application of the bidirectional stack algorithm to speech coding

The performance of the bidirectional metric-first tree encoding algorithm is measured on real speech data coded at 1 bit/s. The main drawback of the unidirectional stack algorithm, variability of its decoding effort as well as decoding erasures, is slightly alleviated. Comparisons, with respect to SNR and time complexity, with the unidirectional stack algorithm and the M-algorithm are made. A procedure for nonlinear bidirectional code design, based on the SGL algorithm is proposed.