ISCA Archive Eurospeech 1993
ISCA Archive Eurospeech 1993

Statistical analysis of the acoustic and prosodic characteristics of different speaking styles

Masanobu Abe, Hirokazu Sato

This paper reports the acoustic and prosodic characteristics of different speaking styles. Three speaking styles are examined by using three different types of texts: a paragraph of an artistic novel, advertisement phrases, and a paragraph of an encyclopaedia. A professional narrator uttered the three texts in appropriate speaking styles that were his own. For convenience, we refer to them as the novel, advertisement and normal speaking style. The analysis results are (1) the 1st formant frequency increases by about 20% in the order of novel, normal, and advertisement speaking style; (2) in terms of the 3rd formant frequency, the novel speaking style is 20% lower in frequency than the other speaking styles; (3) in terms of spectral tilt, the advertisement speaking style has a much flatter spectral tilt than the other speaking styles; (4) F0 range and phrase height assignments are quite different among the three speaking styles; (5) segmental duration in a phrase followed by pause is largely lengthened in the novel speaking style; (6) speech power is commonly modeled as a function of F0 for all speaking styles; and (7) in a syllable followed by pause, vowel devocalization occurs most frequently in the novel speaking style.

Keywords: speaking style, synthesis-by-rule, prosody, formant frequency