Adaptive tree encoding of discrete-time sources with speech applications

Sequentially-adaptive tree encoding procedures are considered for efficient source coding in unknown or time varying situations. Adaptation is performed after a short delay associated with the tree search, and allows convergence to efficient codes for a variety of search complexities, source models, and distortion criteria. Experimental results are presented for a Gauss-Markov source and for speech at 8000 sps, both examined at code rate of one bit per symbol.