ISCA Archive Eurospeech 1993
ISCA Archive Eurospeech 1993

An improvement of the two-level DP matching algorithm using k-NN techniques for acoustic-phonetic decoding

Pablo Aibar, Francisco Casacuberta

The distance-based approach has been used successfully in Isolated and Connected Word Recognition, but this approach has not been explored adequately for Acoustic-Phonetic Decoding. The K-Nearest Neighbor criterion constitutes one of the most powerful classification rules in Pattern Recognition based on the Decision Theory. This rule can be combined with an also powerful but less used algorithm such as the Two-Level algorithm. The main problem with this algorithm was its computational requirements, but this can be reduced adequately. In this paper, we present the use of the combination K-Nearest Neighbor and Two-Level algorithm for Acoustic-Phonetic Decoding. The results that are obtained by this combination are also presented, and they are higher than the ones obtained with more conventional methodologies.

Keywords: Acoustic-Phonetic Decoding, Distance-based approach, Two-Level algorithm, K-Nearest Neighbor rule