Invariance of talkers' choral spectra

The purpose of this study was to determine whether or not amplitude spectra of the “choral” speech [T. Tarnocsy, “Determination du spectra de la parole avec une methode nouvelle,” Acustica 8, 352–358 (1958)] of a talker maintain some relative invariance, irrespective of the language spoken and the occasion the speaker recorded. The algorithm used to make this determination was the HICLUS, which implements Johnson's hierarchical clustering algorithm on the CDC 6500 computer at the Michigan State University [S. C. Johnson, “Hierarchical Clustering Schemes,” Psychometrika 32, 241–254 (1967)]. All pairs of amplitude spectra from a total of 180 were compared, using as a similarity measure the sum of the absolute differences between amplitudes of the same frequencies. In approximately 75% of the cases, the processed data clustered according to subject before clustering according to language or occasion of recording. This suggests that the choral spectra of each speaker have unique characteristics independent of...