On the Normalization of Formant Data

Abstract

The characteristic pattern of a vowel, described by the envelope of its frequency spectrum or by the corresponding loudness pattern on the basilar membrane, is chiefly dependent on the positions of the formant frequencies in the particular scale that is being used. Any vowel may be sufficiently defined by a set of figures related to the formant frequencies. The first three formants for female voices are, on the average, 9 percent higher in the mel scale than the corresponding formants for male voices. Different individuals produce formant patterns of the same relative form on the mel scale. There are several vowels for which significant deviations from the rule of constant mel ratios occur in the comparison of different voices. In these cases, however, the typical feature of the spectrum will be preserved in the relative position of one important formant, or of the mean value of two close formants, which carry the major part of the vowel quality. The mel scale representation is also profitable in the study of the relation between different vowels of a language. Differences of phonemic quality in Swedish vowels are associated with quantal changes of 90 mels in one or more formants.