Modified discrete word recognition

摘要

Speech frames of a user utterance, that includes a plurality of speech elements, are compared to speech models; upon the occurrence of a predetermined condition best candidate speech elements are determined for the utterance by matching speech frames associated with the speech elements to speech models. At times when the speaker separates successive speech elements by pauses longer than a predetermined threshold, discrete speech element recognition is performed on each user utterance; and at other times, more than one speech element is recognized in each user utterance. The user's speech is divided into utterances based on intervening first predetermined conditions appearing between successive utterances; and the utterances are subdivided into speech elements based on a second predetermined condition appearing between successive speech elements. Speech recognition results are delivered based on discrete speech recognition of user utterances including single speech elements and modified discrete speech recognition of user utterances including multiple speech elements.