Dynamic Tree Pruning Method for Fast ACELP Search

The algebraic code excited linear prediction (ACELP) algorithm has been adopted by many speech coding standards, due to low complexity and high quality in its analysis-by-synthesis optimization. One of the optimum proposed algorithms, for reduction of complexity of fix codebook search, is the method based on the pruning of tree with less probability for best excitation in depth first tree search. In this paper, two methods of dynamic pruning of tree, namely single and multi level dynamic tree pruning, are proposed. In addition to low complexity like pruning of tree method, both methods have higher efficiency close to the depth first tree method (one of the most efficient searching methods). The efficiency and complexity of these two methods are compared with pruning of tree and depth first tree methods in an example using by GSM EFR coders.