Discrimination of words in a large vocabulary using phonetic descriptions

摘要: Abstract This paper analyses the results of several experiments performed with aim selecting a suitable representation words for an effective lexical access. A large vocabulary comprising most frequent Italian has been taken into account. Lexical access is in bottom-up phase on basis broad phonetic information, order to reduce number that must be verified. In successive top-down phase, constraints imposed by phonemic structure this set select and schedule (context-dependent) sensory procedures which are appropriate performing detailed phoneme verification analyses, delimited signal intervals, determine, among candidates, word actually spoken. Access lexicon was using different classes descriptions words, ranging from very rough one others quite close form, substantiate relationships between inaccuracy description confusability lexicon. Experiments have both isolated simulate model continuous speech. The obtained orthographic form means translation rules, taking account also possible degradations can occur real system. show that, reasonably feasible acoustic processors, verified reduced, average, about 17 260