Representing Speech Characteristics

Speech characteristics can be roughly described by a few major features: speech flow, loudness, intonation and intensity of overtones. Speech flow describes the speed at which utterances are produced as well as the number and duration of temporary breaks in speaking. Loudness reflects the amount of energy associated with the articulation of utterances and, when regarded as a time-varying quantity, the speaker's dynamic expressiveness. Intonation is the manner of producing utterances with respect to rise and fall in pitch, and leads to tonal shifts in either direction of the speaker's mean vocal pitch. Overtones are the higher tones which faintly accompany a fundamental tone, thus being responsible for the tonal diversity of sounds.

Normative Data

Our approach to analyzing the nonverbal information contained in human speech is based on the results of a normative study on 192 healthy volunteers. The design of this study, with three different types of text and two repeated measurements at 14 day intervals, was chosen to investigate the reproducibility of speech parameters over time, and to analyze the sensitivity of speech parameters with respect to form and content of spoken text. In detail, we determined (1) the optimum recording time required for a reliable estimation of speech parameters, (2) the distribution of speech parameters in the general population, (3) the intra-individual stability of speech parameters over time which allows one to distinguish between "natural" fluctuations and "significant" changes, (4) the differences between dialect and non-dialect, and between affect-neutral and affect-charged speech, (5) the amount of variance explained by the factors age, gender and social status. Within the scope of this normative study we developed a practical recording procedure that can be carried out routinely by a technician in a standardized setting.

Data Analysis

All speech signals are inspected visually and marked with an artifact code if necessary, so that disturbed intervals can be removed prior to data analysis. In a next step, segmentation tables are set up in order to identify pauses and utterances, whereby pauses of less than 250 msec duration are skipped. Subsequently to this, we calculate "spectra" on the basis of 1-second epochs by means of a discrete Fourier transformation ("pure" utterances with pauses having been eliminated for spectral analysis). Finally, we approximate the shape of the F0 distribution curve ("F0" designates the mean vocal pitch of a speaker) by a 2nd degree polynomial and use the distance between the symmetrical 6-dB points as a measure for the "F0-variability" (intonation). The ratio height/width of the 2nd degree polynomial serves as a measure of the "F0-narrowness" (monotony). All frequency differences are calculated in quartertones in order to allow direct comparison between speakers independently of the speakers' mean vocal pitch.

References