Shigeru KATAGIRI, Chin-Hui LEE
A New HMM/LVQ Hybrid Algorithm
for Speech Recognition
Abstract:The Learning Vector Quantization (LVQ) training
algorithms are capable of producing highly discriminative
reference vectors for classifying static patterns. The Hidden
Markov model (HMM) formulation has also been successfully
applied to recognizing dynamic speech patterns. In this paper,
we present a new HMM/LVQ hybrid algorithm for speech
recognition. We show that by combining both the
discriminative power of LVQ and the capability of modeling
temporal variations of an HMM into a hybrid algorithm, the
performance of an HMM-based recognition algorithm is
significantly improved. We tested the hybrid algorithm in a
multi-speaker, isolated word mode, using a highly confusable
vocabulary consisting of the nine English E-set words. The
average word accuracy for the original HMM-based system
was 62%. When the LVQ classifier is incorporated, the word
accuracy increased to 81%.