Czech voiced labiodental continuant discrimination from basic acoustic data

Abstract The Czech voiced labiodental continuant has a special position within the system of Czech consonants. Due to its past it sometimes behaves phonologically as a sonorant, while its canonical articulatory qualities place it among obstruents. It is quite difficult to decide how much of its sound is sonorant and how much is obstruent by mere auditory analysis. We investigated whether basic acoustic parameters extracted from fluent reading material can help to discriminate if the Czech [v] leans more towards the obstruent or the sonorant class of consonants. Dynamic profile data were processed by factor analysis and, together with duration and harmonicity measurements, used as an input for cluster analysis, discriminant analysis, and classification-tree analysis. The results show that acoustically, the Czech intervocalic [v] in connected speech is more in line with its sonorant past rather than its obstruent presence. 1. Introduction The consonantal inventory of modern Czech consists of two natural classes - obstruents and sonorants. The obstruents form pairs where specific place and manner of articulation is represented by one voiceless and one voiced segment. In certain positions these segments trigger or undergo assimilation of voicing. The sonorants, on the other hand, do not come in pairs, they are always voiced, and they neither trigger nor undergo assimilation of voicing. One exception to this neat arrangement exists, however. The labiodental pair /f, v/ is formed by a voiceless /f/ which behaves as a proper obstruent in that it both triggers and undergoes assimilation, and a voiced /v/ which only undergoes but does not trigger assimilation of voicing.For explanation of this ambiguity one has to look back to the history of the language. It is well established that Old Czech did not have any labiodental sounds. In late Middle Ages, /f/ was imported into the language through foreign words, and formed a considerably asymmetrical pair with the already present bilabial approximant /w/. The pull towards greater symmetricity within the system initiated a gradual change of /w/ to /v/. Despite the fact that this change has been going on for over five centuries [1], it has not been completed. Unfortunately, the lack of accurate data prevents us from examining its course. We do not know how fast or steady the process has been and we cannot appreciate the mutual relationship of phonetic and phonological aspects of the change. As indicated above, Czech /v/ keeps some phonological characteristics of a sonorant. As to its phonetic make-up, however, the literature of the past century or so has occasionally questioned its fricativeness on behalf of plosiveness (e.g., [2]), but never put forward the suggestion that the sound may have retained some of its sonorant quality. Since our informal observations suggested that [v] in contemporary speech is not as unambiguous an obstruent as indicated in literature, we decided to put its phonetic properties to test. The question was whether basic acoustic data fed into a recognition algorithm would place [v] closer to another unequivocal sonorant or obstruent sound. To verify the discriminatory power of our analyses, we also included a pair of related voiceless fricatives, and to address the question of occlusion and release, a plosive sound was added. Given that we had over 1500 phones uttered by 265 different speakers and the data fed into the recognition algorithms were constrained, it was clear that the recognition rate would be short of 100 %. The resulting errors, then, should provide the answer to our question. If [v] is misclassified as one of the obstruents more often than as the sonorant, then the literature is right in calling it an obstruent without reservations. If, however, it is misclassified as the sonorant more often than as one of the other obstruents, the view of contemporary Czech labiodental continuant should be modified.