First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

The Speech Code. Segmental and Prosodic Features

Gunnar Fant

Department of Speech Communication and Music Acoustics, Royal Institute of Technology (KTH), Stockholm, Sweden

This is a review of basic issues in speech analysis with an emphasis on work carried out at the KTH in Stockholm. The concept of the speech code as a knowledge base for speech research and applications is outlined. In a narrow sense the speech code may be substantiated by programs and rules for text-to-speech conversion, in a wider sense also covering speaker type and speaking style and other aspects of a message not inherent in a text. For adequate descriptions we need the support of extensive speech analysis, articulatory modelling and perceptual critera. This is exemplified by a discussion of labial and velar stops in a back vowel context. One aspect of the interaction between prosodic and segmental features is the degree of articulatory and associated spectral contrast which becomes enhanced with increasing stress. A brief review is made of stress correlates based on objective and subjective measures with comments are made on language differences. Interstress intervals do not display physical isochrony but some readers plan their pauses to integrate in rhythmical synchrony with a perceived local average stress rate, The average pause-to-speech ratio is a speaker specific characteristic.

Full Paper

Bibliographic reference.  Fant, Gunnar (1990): "The speech code. segmental and prosodic features", In ICSLP-1990, 1389-1398.