Evaluating multiple LVCSR model combination in NTCIR-3 speech-driven web retrieval task

@inproceedings{Matsushita2003EvaluatingML,
  title={Evaluating multiple LVCSR model combination in NTCIR-3 speech-driven web retrieval task},
  author={Masahiko Matsushita and Hiromitsu Nishizaki and Takehito Utsuro and Yasuhiro Kodama and Seiichi Nakagawa},
  booktitle={INTERSPEECH},
  year={2003}
}
This paperstudiesspeech-dri ven Web retrieval modelswhich acceptsspokensearchtopics (queries)in the NTCIR-3 Web retrieval task. The major focusof this paperis on improving speechrecognitionaccuracy of spokenqueriesand then improving retrieval accuracy in speech-dri ven Webretrieval. We experimentallyevaluatethetechniquesof combiningoutputsof multiple LVCSR modelsin recognitionof spokenqueries.As modelcombinationtechniques, we comparetheSVM learning techniqueand conventionalvoting schemessuchas… CONTINUE READING