摘要 |
<p>Speech recognition apparatus has a vocabulary store 2 containing pattern templates or other acoustic information, grammar and collocation information about each word that can be recognised. An acoustic recogniser 30 compares the output of a microphone 1 with the pattern templates in the store 2 to produce signals representing the most likely words spoken and their grammar and collocation. The grammar information is used to check that the words identified by the recogniser are compatible with each other. Grammar information can also be used to restrict the words in the store with which following words are compared by the recogniser 30. A semantics unit 40 compares collocation information of identified words with other identified words in a phrase to check that they are compatible. If identified words are incompatible in grammar or collocation, alternative, less likely words are selected for comparison. When a word is identified with sufficient probability that is compatible in grammar and collocation, an output is generated to a utilisation device 5. If no word is identified with sufficient certainty, a feedback signal is produced to prompt repetition of the word.</p> |