ESCA Tutorial and Research Workshop on
The need for a comprehensive, standardized speech database is threefold: first, to acquire acoustic-phonetic knowledge for phonetic recognition; second, to provide speech for training recognizers; and third, to provide a common test base for the evaluation of recognizers. There are many factors to consider in corpus design, making it impossible to provide a complete database for all potential users. It is possible, however, to provide an acceptable database that can be extended to meet future needs. After much discussion among several sites, a consensus was reached that the initial acoustic-phonetic corpus should consist of calibration sentences, a set of phonetically compact sentences, and a large number of randomly selected sentences to provide contextual variation. The database design has been a joint effort including MIT, SRI, and TL This paper describes MIT's role in corpus development and analyses of the phonetic coverage of the complete database. We also include a description of the phonetic transcription and alignment procedure.
Bibliographic reference. Lamel, Lori F. / Kassel, Robert H. / Seneff, Stephanie (1989): "Speech database development: design and analysis of the acoustic-phonetic corpus", In SIOA-1989, Vol.2, 161-170.