Re: National differences in the tritone paradox (at)

Subject: Re: National differences in the tritone paradox
From:    at <parncuttSOUND.MUSIC.MCGILL.CA>
Date:    Sat, 30 Oct 1993 01:26:35 EDT

The saga continued: How to explain variations in the main pitch of an OCT as a function of language or dialect? The main pitch of an OCT, according to TSS82, is the most salient of a set of virtual pitches. The salience of each virtual pitch depends on the saliences of spectral pitches that are quasi-harmonics of that virtual pitch. The salience of those spectral pitches in turn depends on their absolute frequency: Spectral pitches falling in a specific range, called the spectral dominance region, tend to be more salient, or tend to have more influence on virtual pitches, than do pitches outside that range. The spectral dominance region is formulated in TSS82 by a "spectral frequency weight" -- a mound-shaped function that is symmetrical wrt log frequency, with a broad peak at 700 Hz. The function reflects the average relative perceptual importance of partials of speech vowels, as a function of their frequency. The spectral frequency weight function, like just about everything else in TSS82 (and most of psychophysics, for that matter) is an AVERAGE that applies APPROXIMATELY to a given population of listeners. An intriguing possibility is that the exact center and/or shape of this function may depend on dialect or language. It is clear that the timbre, and hence spectral envelope, of specific vowels differs from one accent, dialect, or language to another. For example, an American "oh" is timbrally deeper than an English "oh", suggesting that the American "oh" has a lower centroid frequency. Might American vowels be generally (or, at least, on average) "timbrally deeper" than English ones? Has anyone measured long-time average spectra of the speech of different languages and dialects to find out if there are consistent differences in the distributions? If such differences exist, then the shape and center of the spectral dominance region in TSS82 may vary as a function of language or dialect. That could explain variations in the pitch distribution of OCTs, and thereby explain the observed variations in the "tritone paradox" phenomenon. Richard Parncutt

This message came from the mail archive
maintained by:
DAn Ellis <>
Electrical Engineering Dept., Columbia University