Phonetic properties of a large Spanish lexicon and its implications for large vocabulary speech recognition

One way to improve the efficiency of isolated-word recognition systems for large vocabularies is the use of broad acoustic classes to prune the lexicon to a manageable size and even to recognize uniquely a considerable part of the lexicon itself based on phonotactic constraints. A preliminary study of two different Spanish lexicons has been made to determine the adequateness of such an approach for Spanish. Preliminary results show a similar behavior compared to English.<<ETX>>