Using quartiles of F1 in the identification of variants of /v/ in Romanian and French

We present a method for identifying the allophonic variants of /v/ using the fundamental frequency and the first and the third quartiles of the first formant. We apply the method to French and Romanian and compare the use of the allophones. We obtained two classes for the voiced (/v_v/) and partly voiced (/v_vo/) /v/ and a single class for the unvoiced /v/ (/v_o/) for both languages. We determine the conditional probabilities of the /v/ classes. The results show a high similarity between the two languages and that Q1 and Q3 are relevant statistical parameters for the /v/ clustering. We discuss the possibility of using the conditional probabilities of the /v/ classes in dental screening.