Speech Recognition Using a Discriminative, Context-Independent, Segment-Based Speech Recognizer
Jan Verhasselt, Jean‐Pierre Martens, Baeyens Bart · Ghent University Academic Bibliography (Ghent University) · 1996
In this paper, we describe important improvements that were recently introduced in our Discriminative Stochastic Segment Model (DSSM) speech recognizer. We propose a new presegmentation algorithm and we optimize the structure of the Multi-Layer Perceptron (MLP) that estimates the phone probabilities. Additionally, we describe a cascade MLP combination technique that relaxes the drawbacks of traditional stochastic segment models. The proposed improvements have resulted in a statistically significant increase of the speaker-independent continuous phone recognition performance on the TIMIT corpus. Keywords--- Segment-Based Speech Recognition, Multi-Layer Perceptron, Presegmentation, Multiple Classifier Combination I. Introduction Our Discriminative Stochastic Segment Model (DSSM) continuous speech recognition system [1], [2], incorporates an auditory model front-end [3], a presegmentation algorithm, Multi-Layer Perceptrons (MLP's) for phonetic classification and segmentation, a lexical ...