Automatic classification of consonant clusters in French

Abstract

This study describes a method for the automatic detection of two-consonant clusters in French. Four corpora were used, consisting of 603 different consonant clusters and single consonant combined with the 3 vowels /i, a, ã/, in disyllabic and trisyllabic words. The CCV, VCC, CV and VC syllable structures were studied. Stimuli were recorded 5 times in an anechoic room by 10 subjects. Word length, syllable length and syllable-component length were measured by means of a signal editor. A set of rules was deduced from a statistical analysis carried out on these data. Five relevant parameters were extracted, namely (1) voice feature, (2) mode of articulation in the first half of the cluster, (3) duration ratio between the vowel and consonant segments, (4) duration of the consonant segments, and (5) position (prevocalic, intervocalic and postvocalic). This phonetic approach was examined on the GRECO-BDSONS public test corpus. More than 90% of the consonant clusters were correctly classified on the basis of the values of the duration parameters extracted from our data base. Seventeen rules were used to output all the macro-classes of the consonant clusters in the test corpus.