Multilayer perceptron postprocessor to Hidden Markov modeling for speech recognition

Abstract

In this paper, a new neural network postprocessor is introduced to enhance the classification capability of hidden Markov modeling for speech recognition. This postprocessor receives stimuli from not one but all word- HMMs for each word speech and does not require segmenting speech frames at subword level. A multilayer perceptron implementation has achieved 20% to 30% syllable error reduction in experiments reported here.