[′kē‚wərd ‚spät·iŋ] (engineering acoustics) An approach to task-oriented speech understanding through detecting a limited number of keywords that would most likely express the intent of a speaker, rather than attempting to recognize every word in an utterance.