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

Banff, Alberta, Canada
October 13-16, 1992

Generation and Extraction of High Quality Synthesis Units

J. R. Andrews (1), K. M. Curtis (1), Volker Kraft (2)

(1) Department of Electrical and Electronic Engineering, University of Nottingham, UK
(2) Lehrstuhl f. Allgemeine Elektrotechnik u. Akustik, Ruhr-Universitšt Bochum, Germany

This paper proposes a novel method for the generation and extraction of high quality speech synthesis units, namely demisyllables. The design incorporates a hybrid system combining both formant speech analysis and synthesis, enabling the extraction and subsequent formant coding of the demisyllables. The German language was used as a source for the demisyllable extraction. The identification of the speech synthesis units was based upon the analysis of data obtained from three different databases for the chosen language. The demisyllables were embedded in carrier words in specifically constructed sentences, coded in SAMPA and recorded in a sterile environment. SAMPA having been chosen as it is the new standard European phonetic alphabet allowing multi-lingual transcription. This approach ensured high speech quality and uniformity in the synthesis elements. The analysis and synthesis system was designed and implemented on a dual processing structures thus resulting in maximum system performance. We present the results of the embedment, extraction, recording and formant coding techniques.

Full Paper

Bibliographic reference.  Andrews, J. R. / Curtis, K. M. / Kraft, Volker (1992): "Generation and extraction of high quality synthesis units", In ICSLP-1992, 1203-1206.