Sixth European Conference on Speech Communication and Technology
(EUROSPEECH'99)

Budapest, Hungary
September 5-9, 1999

Towards Recognizing "Non-Lexical" Words in Spontaneous Conversational Speech

Hesham Tolba, Douglas O'Shaughnessy

Institut National de la Recherche Scientifique, INRS-Télécommunications, Québec, Canada

The purpose of this paper is to study and analyze both the -\it non-lexical- filled pauses and intended responses in conversational spontaneous speech, and how this can be useful in both automatic speech recognition and speaker identification systems. Through experiments, it was found that we are able to distinguish between words and non-lexical words in spontaneous speech using prosodic features. Consequently, a pre-recognition of such pauses using a decision-tree based CART classifier is evident. Thus, for ASR of spontaneous speech, such pauses can be either totally omitted or considered as words to be added to the dictionary of the ASR system, consequently improving the performance of such an ASR system.


Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Tolba, Hesham / O'Shaughnessy, Douglas (1999): "Towards recognizing "non-lexical" words in spontaneous conversational speech", In EUROSPEECH'99, 723-726.