Paper: An Empirical Study On Multiple LVCSR Model Combination By Machine Learning

ACL ID N04-4004
Title An Empirical Study On Multiple LVCSR Model Combination By Machine Learning
Venue Human Language Technologies
Session Short Paper
Year 2004
Authors

This paper proposes to apply machine learn- ing techniques to the task of combining out- puts of multiple LVCSR models. The proposed technique has advantages over that by voting schemes such as ROVER, especially when the majority of participating models are not reli- able. In this machine learning framework, as features of machine learning, information such as the model IDs which output the hypothe- sized word are useful for improving the word recognition rate. Experimental results show that the combination results achieve a relative word error reduction of up to 39 % against the best performing single model and that of up to 23 % against ROVER. We further empirically show that it performs better when LVCSR mod- els to be combined are chosen so as to cover as many correctly recognized word...