Second ESCA/IEEE Workshop on Speech Synthesis

September 12-15, 1994
Mohonk Mountain House, New Paltz, NY, USA

A Structured Way of Looking at the Performance of Text-To-Speech Systems

Louis C.W. Pols (1), Ute Jekosch (2)

(1) Institute of Phonetic Sciences / IFOTT, University of Amsterdam, The Netherlands
(2) Lehrstuhl fuer Allgemeine Elektrotechnik & Akustik, Ruhr-Universitaet Bochum, Germany

Via the Cocosda Bulletin Board an extensive QUESTIONNAIRE was distributed in 1993, upon which 16 reactions have been received. Almost all these reactions were very interesting and detailed. They contain a wealth of ideas and suggestions. In our presentation for this ESCA/IEEE Workshop we combine, on the one hand the suggestions from the respondents with, on the other hand the knowledge and experience collected in various projects (SPIN, SAM(-A), Eagles, and (Euro)Cocosda), into a new approach for a structured way of evaluating the performance of text-to-speech systems. The basic idea is to define a set of keywords/descriptors that specify the system under study, with special emphasis on its application. In a similar way the available and to-be-developed tests should be characterized. System/application can then be linked, in a matrix-type way, to the suite of tests, and a proper selection can then be made, or it might become apparent that additional specific tests are still required.

Full Paper

Bibliographic reference.  Pols, Louis C.W. / Jekosch, Ute (1994): "A structured way of looking at the performance of text-to-speech systems", In SSW2-1994, 203-206.