Third International Conference on Spoken Language Processing (ICSLP 94)
We are developing a speaker independent speech recognition method using the similarity vectors as feature parameters. In this paper we present the feasibility of the technique in practical usage by downsizing the algorithm and improving the word spotting technique. When tested with 100 Japanese city names in noisy environments, a word recognition rate 95.9% was obtained with reduced memory size and computation amount. We expect that the method can be implemented in a small hardware.
Bibliographic reference. Hoshimi, Masakatsu / Yamada, Maki / Niyada, Katsuyuki (1994): "Speaker independent speech recognition method using phoneme similarity vector", In ICSLP-1994, 1915-1918.