Efficient A* Stack Decoder Algorithm for Continuous Speech Recognition with a Stochastic Language Model.

摘要

The stack decoder is an attractive algorithm for controlling the acoustic and language model matching in a continuous speech recognizer. It implements a best-first tree search to find the best match to both the language model and the observed speech. A previous paper described a near-optimal admissible Viterbi A* search algorithms for use with non-cross-word acoustic models and no-grammar language models(1). This report extends this algorithm to include unigram language models and describes a modified version of the algorithm which includes the full (forward) decoder, corss-word acoustic models and longer-span language models. The resultant algorithm is not admissible, but has been demonstrated to be very efficient.