Pitch analysis methods for cross-speaker comparison

A system of fundamental frequency (F0) analysis and normalisation is described for obtaining pitch data and comparison across speakers. This system was used for the analysis of English and Spanish speakers' productions in order to compare the realization of accentual focus in the two languages. The system is based on the simultaneous recording of speech and the laryngeal signal. The latter is monitored by means of an electrolaryngograph. The analysis is done in three stages: (a) auditory analysis; (b) analysis using the PCLX software package from Larynograph Ltd., obtaining measures in Hz for peaks, troughs and other relevant points in the contour, and also fundamental frequency statistics from the complete data set for each speaker; and (c) a detailed analysis using the Speech Filing System (SFS) with simultaneous display of speech pressure waveform, laryngograph waveform (Lx), excitation period measurements and F0 trace, and with playback facilities for speech and F0. The analysis described for (c) is used for segmentation, to filter out nodes which are due to micro-intonation, and to pinpoint problem areas in the F0 trace. Outlying and anomalous period measurements may be replaced by a five-point median value. The resulting contours are normalised by converting Hz measures to percentage values (positive or negative) of the speaker's mean F0, which is obtained from the analysis described for (b).