Abstract
We describe an improvement of the design of the phoneme recognizer that is based on the articulatory feature (AF). Several strategies for designing the optimal parameter set in AF-based Hidden Markov Model (HMM) are investigated. They include subword units, number of HMM states, vowel group separation, tuned insertion penalty, and HMM topologies. The proposed AF-based phoneme recognition with 5-state HMMs, separated vowel, triphone subword, Bakis topology, and optimal insertion penalty provides the best accuracy among the experiments, i.e., 81.38% for the JNAS speech database. This result surpasses the accuracy of the standard MFCC-based phoneme recognition for triphone subword, 3-state HMMs, and 16 Gaussian mixtures.
