Next: The Geometrical Processing Up: Projects Previous: Study of Machine

Study on Speech Recognition under Noisy Environment

Masahide Sugiyama,
Professor, Human Interface Laboratory

We implemented server/client type FPM-LR speech recognizer to achieve high speed and parallel processing. The client sends speech wave to each recognition server, and receives the results of feature extraction, recognition process and the phrase recognition, and constructs the final sentence candidates. On the microphone with range-dependent sensitivity, a 41 unit microphone achieved the almost upper bound of analogue processing, then to get high performance new digital processing based and practical 17 unit microphone is now being developed.


www@u-aizu.ac.jp
January 1996