Title :
Some experiments with a supervised classification procedure
Author_Institution :
Continuous Speech Recognition Group, IBM Thomas J. Watson Res. Center, Yorktown Heights, NY, USA
fDate :
7/1/1994 12:00:00 AM
Abstract :
A supervised classifier is described where a systematic search through a training database produced an efficacious set of prototypes for its operation. Using this technique, two studies, one dealing with phone recognition and another with large-vocabulary speech recognition, were conducted with data from several speakers. The superiority of the supervised classifier design is demonstrated by comparative performance analysis
Keywords :
speech recognition; telephony; experiments; large-vocabulary speech recognition; performance analysis; phone recognition; speaker-dependent speech recognition; supervised classification; supervised classifier; systematic search; training database; Databases; Hidden Markov models; Performance analysis; Prototypes; Sampling methods; Solids; Speech recognition; Training data; Viterbi algorithm;
Journal_Title :
Speech and Audio Processing, IEEE Transactions on
