Prosodic Boundary Detection

Abstract

Prosodic constituent structure, or the perceived grouping of words in speech, plays a role in human speech communication in virtually every language. Speakers use prosodic phrasing to contribute meaning to and sometimes disambiguate the sequence of words that comprise an utterance by highlighting its information structure. From a speech analysis perspective, prosodic phrase structure provides the link that seems to most effectively explain continuously varying acoustic correlates (pauses, FO patterns, duration lengthening, etc.) in terms of the word sequence of an utterance (syntactic, semantic and discourse structure). Just as both speakers and listeners use prosodic phrases in human speech communication, so computational models of prosodic phrase structure can be useful both for communicating meaning in synthesized speech and for extracting meaning in automatic speech understanding. In fact, prosodic phrase structure is probably even more important for computer speech processing than for humans, because computers have a much less detailed semantic representation and less extensive knowledge of the world than humans and thus word sequences tend to be more often ambiguous in computer language processing than for human listeners.KeywordsSpeech RecognitionSpontaneous SpeechWord SequencePitch AccentPhrase BoundaryThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.