Uma Nova Abordagem Fonetico-Fonologica em Sistemas de Reconhecimento de Fala Espontânea DOI: 10.14209/jcis.2003.13

This work describes a continuous speech recognition system. using a phonetic-phonological approach, covering a large vocabulary. The basic theory and implementation technique for the acoustic pattern recognition is the well-known Hidden Markov Models (HMM). Furthermore, the linguistic approach is enhanced, spotting the concepts, definitions and research lines in Linguistics, providing useful and applicable elements to build a large vocabulary speech recognition system. This system improves the speech recognition performance with phoneticphonological modeling. The general likelihood scores are increased, getting a better recognition performance, due to the statistical phonetic structure enhancement of some frequent phonetic combinations usually found in the idiom. The utilization of this new phonetic-phonological approach is strongly recommended in spontaneous speech recognition systems.