Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

Extrinsic Normalization of Vowel Formant Values Based on Cardinal Vowels Mapping

Maria-Gabriella Di Benedetto (1), Jean-Sylvain Lienard (2)

(1) INFOCOM Dept, Facolta di Ingegneria, University of Rome 'La Sapienza', Roma, Italia
(2) LIMSI-CNRS, Orsay, France

An extrinsic vowel normalization method is proposed. In the method proposed, a-priori information referring to the first and second formant values (Fl and F2) of three vowels of each speaker is used. In order to obtain the normalized formant values of a given speaker's vowels in a given language it is necessary to:

  1. gather reference information on three vowels, for example [i,#,u]. This information corresponds to the Fl and F2 values for [i,map the formant values of [i,a,u] of the new speaker, onto the reference formant values. This operation can be achieved by linear transformation (isomorphic transformation).
  2. the normalized formant values of the other vowels can be obtained by applying the isomorphic transformation. Results are presented for vowels of American English, and briefly reported for vowels of Italian, and French. Comparison between a classical F1-F2 and the mapped vowel representations in terms of statistical parameters are discussed.

Full Paper

Bibliographic reference.  Benedetto, Maria-Gabriella Di / Lienard, Jean-Sylvain (1992): "Extrinsic normalization of vowel formant values based on cardinal vowels mapping", In ICSLP-1992, 579-582.