This paper describes the acoustic processor of a Spanish Continuous Speech Recognition System based on HMM Demisyllable units. The acoustic processor produces a lattice of word hypotheses suitable to be parsed by a linguistic analyzer and it is based on a spotting algorithm. This spotting algorithm has three inputs : The unknown utterance, the HMM of the dernisyllables or references and the lexical knowledge in terms of a finite state network . In this paper , the authors describe a less cost method for speaker independent continuous speech Word Spotting.This spotting algorithm is a modified version of the OneStep Viterbi algorithm with multiple hypothesis. The implemented acoustic processor has been tested using telephonic numbers and integers numbers from Oto 1000 in a speakerindependent context: Excellent results have been obtained that have demostrated the efficiency of the spotting algorithm and the performance of the demisyllable as recognition unit.

Peer Reviewed