Abstract | ||
---|---|---|
This paper proposes to apply machine learning techniques to the task of combining outputs 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 reliable. In this machine learning framework, as features of machine learning, information such as the model IDs which output the hypothesized 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 models to be combined are chosen so as to cover as many correctly recognized words as possible, rather than choosing models in descending order of their word correct rates. |
Year | Venue | Keywords |
---|---|---|
2004 | HLT-NAACL (Short Papers) | word correct rate,single model,word recognition rate,relative word error reduction,empirically show,multiple lvcsr model combination,model ids,machine learning,lvcsr model,multiple lvcsr model,empirical study,combination result |
Field | DocType | ISBN |
Voting,Computer science,Word recognition,Speech recognition,Artificial intelligence,Natural language processing,Empirical research,Machine learning | Conference | 1-932432-24-8 |
Citations | PageRank | References |
6 | 0.92 | 7 |
Authors | ||
5 |
Name | Order | Citations | PageRank |
---|---|---|---|
takehito utsuro | 1 | 456 | 82.76 |
Yasuhiro Kodama | 2 | 19 | 2.87 |
Tomohiro Watanabe | 3 | 13 | 2.20 |
Hiromitsu Nishizaki | 4 | 163 | 29.49 |
Seiichi Nakagawa | 5 | 598 | 104.03 |