Data Processing for Analyzing Intonational Signals of American English
Yukio Takefuta, Donald M. Morehead · The Journal of the Acoustical Society of America · 1970
Eleven different methods for processing the measures of F0 on the melody curves of American English were evaluated with regard to their relative efficiencies in detecting intonational signals. The processing methods to be evaluated were chosen on the basis of their ability to analyze the measures of F0 and reduce them to a single digit to be related with any one of such psychological sensations as pitch pattern, pitch level, rate of pitch change, range of pitch change, and duration of pitch. A method of combining a linear regression analysis with an appropriate segmentation was introduced in this study to analyze a complex melody pattern into a set of simple linear patterns. Relative efficiencies of all the processing methods were evaluated by computing the classical indexes of signal detectability. The method introduced to quantize the melody patterns of utterances was the most efficient of all the processing methods tested for detecting the intonational signals. [This research was supported by a contract between The Ohio State University Research Foundation and the Office of Naval Research.]