Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
An important problem in the practical use of speech recognition technology is the robustness to environment noise and unnecessary utterances. We have already reported in IEEE ICASSP '87[i] that the recognition method using the word spotting technique is effective on such conditions. However, the word spotting error, which part of a word incorrectly matches with other words occurred in some vocabularies. CLM-alpha(Improved Continuous Linear Expansion / Compression Matching) Method was adopted as a recognition method. In this method, two matching procedure steps are carried out; a primary matching procedure is characterized by word spotting, and a secondary matching procedure is performed on the sequence of similarity scores obtained through the primary matching procedure. As a result, the word spotting error was minimized and the overall recognition performance was improved.
In addition, the application of newly developed voice-recognition LSI which uses a CLM-alpha method in the remote control for VCR has made it possible to make program reservations using voice dialog. In an evaluation test using the hardware implementation of the control, an average recognition rate of 95.9% was obtained with a 58-word vocabulary which was divided into 6 word groups, spoken by 141 men and women in an office environment. At present, the Japanese version is on the market.
Bibliographic reference. Kuwano, Hiroyasu / Nomura, Kazuya / Ookumo, Atsushi / Hiraoka, Shoji / Watanabe, Taisuke / Niyada, Katsuyuki (1992): "Speaker independent speech recognition method using word spotting technique and its application to VCR programming", In ICSLP-1992, 1527-1530.