European Conference on Speech Technology

Edinburgh, Scotland, UK
September 1987

Speech Analysis and Synthesis Methods Based on Spectral Envelopes and Voiced/Unvoiced Functions

Xaviefr Rodet (1,2), P. Depalle (1,3), G. Poirot (3)

(1) ERCAM, Paris, France
(2) L.A.F.O.R.I.A., University Paris, France
(3) Ecole Superieure d'Electricité, Metz, France

The Institut de Recherche et de Communication Acoustique/Musique (IRCAM) is involved in sound analysis and synthesis for contemporary Music Creation. Accurate analysis and high quality synthesis of sounds are thus extensively studied. We describe here some of the methods that have been developped for speech and singing voice. They can be used for accurate analysis, for sound processing (such as time or spectral warping, filtering, pitch transposition, etc..) and for synthesis-by-rule. The goal is also to provide powerful tools and to get sounds with as high a quality as possible since they will be used in pieces and concerts, but speech applications are equally in view (4). At present, speech signals are digitized at 16KHz. on 16 bits. Higher sampling rates are considered.

Full Paper

Bibliographic reference.  Rodet, Xaviefr / Depalle, P. / Poirot, G. (1987): "Speech analysis and synthesis methods based on spectral envelopes and voiced/unvoiced functions", In ECST-1987, 1155-1158.