Viterbi beam search with layered bigrams

Minimizing search errors due to delayed bigrams in real-time speech recognition systems

When building applications from large vocabulary speech recognition systems, a certain amount of search errors due to pruning often has to be accepted in order to obtain the required speed. In this paper we tackle the problems resulting from aggressive pruning strategies as typically applied in large vocabulary systems to achieve close to real-time performance. We consider a typical scenario of...

KL-Divergence Guided Two-Beam Viterbi Algorithm on Factorial HMMs

This thesis addresses the problem of the high computation complexity issue that arises when decoding hidden Markov models (HMMs) with a large number of states. A novel approach, the two-beam Viterbi, with an extra forward beam, for decoding HMMs is implemented on a system that uses factorial HMM to simultaneously recognize a pair of isolated digits on one audio channel. The two-beam Viterbi alg...

Backward Viterbi beam search for utilizing dynamic task complexity information

The backward Viterbi beam search has not received enough attentions other than being used in the second pass. The reason is that the speech recognition society has long ignored the concept of dynamic complexities of a speech recognition task which can help us to determine whether we should operate Viterbi decoding in forward or backward direction. We use the U.S. street address entry task as on...

Various Robust Search Methods in a Hungarian Speech Recognition System

This work focuses on the search aspect of speech recognition. We describe some standard algorithms such as stack decoding, multi-stack decoding, the Viterbi beam search and an A* heuristic, then present improvements on these search methods. Finally we compare the performance of each algorithm, grading them according to their performance. We will show that our improvements can outperform the sta...

s 2 3 s / A D / A D Noise Signal Beam - Search Viterbi Models Initial Extractor