Erik MCDERMOTT and Shigeru KATAGIRI
LVQ-Based Shift-Tolerant Phoneme Recognition
Abstract:In this paper we describe a shift-tolerant neural network architecture for
phoneme recognition. Our system is based on algorithms for LVQ (Learning
Vector Quantization) [1,2], recently developed by Teuvo Kohonen, which pay
close attention to approximating optimal decision lines in a discrimination
task. Recognition performances in the 98-99% correct range were obtained for
LVQ networks aimed at speaker-dependent recognition of phonemes in small
but ambiguous Japanese phonemic classes. A correct recognition rate of 97.7%
was achieved by a single, larger LVQ network covering all Japanese consonants.
These recognition results are at least as high as those obtained in the Time
Delay Neural Network system developed by Alex Waibel et al. (1988), and
suggest that LVQ could be the basis for a successful speech recognition system.