Combining speech feature information by D-S evidence reasoning
Wang Chengyou, Liu Yunhui, Cai Xuan-ping, Liang Dian-nong · 2004
D-S evidence reasoning is applied to speech recognition for combining various kinds of speech feature information in this paper. The principle of combining feature information with D-S evidence reasoning is firstly described. Then the paper discusses the decision rules for recognition and computation of the method. An experiment is finally finished in a DHMM/MNCD recognition system using the method and the experiment results are discussed. It improves about 5% in the accuracy rate of the experiment system, which speech base has 100 Chinese similar words in pronunciation. The results indicate that D-S evidence reasoning could combine feature complementary information in speech recognition for improving recognition performance.