7th International Conference on Spoken Language Processing

September 16-20, 2002
Denver, Colorado, USA

Acoustical Correlates to SD Ratings of Speaker Characteristics in Two Speaking Styles

Yasuki Yamashita (1), Hiroshi Matsumoto (2)

(1) Nagano Prefectural Institute of Technology, Japan; (2) Shinshu University, Japan

For synthesizing voice quality expressed by adjectives, this paper investigates acoustical correlates to adjective ratings of speaker characteristics for reading and conversational speech. The results revealed: (1) The speaking styles have little effect on the rates on adjective scales. (2) The effects of formant frequencies and long-term spectrum to adjective ratings are almost independent of speaking styles. (3) The "busy" voices have a significant correlation to the standard deviation of logarithmic F0 divided by speech rate (mora per second) for both reading and conversational voices. Furthermore, for conversational speech, the "busy" ratings significantly correlate to the F0 as well. (4) For the reading speech, "articulate" voices have larger dispersion on the F1-F2 plane than "inarticulate" ones. For conversational speech, both the "articulate" and "inarticulate" voices have small dispersions compared to those for reading speech.


Full Paper

Bibliographic reference.  Yamashita, Yasuki / Matsumoto, Hiroshi (2002): "Acoustical correlates to SD ratings of speaker characteristics in two speaking styles", In ICSLP-2002, 2577-2580.