Discriminant Analysis of French Stops and Nasals.

Abstract

Studies on the invariant features of Japanese stop and nasal consonants have been extended to French. Place and or manner of articulation of 7 consonants/ʔ, p, t, k, b, d, g/ are discriminated in an environment of /a, o, œ, e, ɛ, u, y, i, ã, ɛ, õ/. The feature vector is 23 LPC cepstrum coefficients at every 10 ms of the initial 100 ms (30 ms before the burst and 70 ms after the burst). The burst point was manually specified referring to waveform display. Three of nasal consonants /m, n, ɲ/ are discriminated in the same environment. The release point, openning of the oral passage, was manually specified also. The stepwise discriminant analysis in the SAS system obtained a reduced feature set and discriminant score. The 40 male speakers' sample comprizes 3080 stops and 1120 nasals. Speaker and vowel independently stops are better discriminated than Japanese, but nasals are equivalent or a little lower than Japanese. The conclusion that the spectral pattern near the stop burst is a good feature for place discrimination can be generalized throughout French and Japanese. Results are compared with perception test.