Volume 33, Issue 6

Determining whether or not two arbitrary vowel tokens are phonetically equivalent is the crucial step in classifying vowels phonetically. Acoustic phonetic equivalence for vowel sounds is discussed in the light of data on formant frequency variability. Random variation between repetitions of the same item by the same speaker or between imitations of the same item by different speakers determines a range of practical limits on the sizes of acoustic phonetic equivalence classes. Vowel imitation is suggested as an approach to defining vowel equivalence across speakers. Effects of prosody and context can be handled cither by estimating vowel targets or by permitting an intended or perceived vowel type to have a number of condition-dependent variants. The consequences of these options are discussed in terms of their practicality and consistency with basic phonetic concepts.

Johansson’s experiments reveal that /d/ is identified mostly thanks to its transitions, in opposition to /b/ and /g/, a consideration which sustains the author’s opinion. By means of EVA III, he first synthesized /da/ and /du/ after DELATTREs rules, then after natural patterns. The latter turn out to be quite more intelligible, a consideration which shows that the extrapolated ‘loci’ which Delattre had to use in order to avoid categorization confusions, are not able to ensure ‘maximal’ consonantic perception; his experiments must have been altered by confusion between ‘transitions’ and ‘sonorous attacks’. The author considers logatoms as ‘aggregate patterns’, tackles formantic anamorphosis and ‘balance’, and gives notice of the realization of an ‘evagraphic dictionary’ intended for computer speech.