Spoken Language Assessment in the European Context

The increasing scope of work in the field of Speech and Linguistic Engineering is gradually leading to the future introduction of robust techniques for real-time multilingual operation. The ‘Speech Assessment Methods’ ESPRIT Project 2589 (SAM) is directed towards the definition and application of multilingual standards to assist in the assessment, evaluation, cross-comparision, and further advance of these developments. The SAM group involves collaboration between twenty-six laboratories in eight countries, SIX within the EEC and two from EFTA. Our work is concerned with the establishment of databases, speech and language descriptive methods, and quantitative tools for the assessment both of speech recognizers (‘input’) and speech synthesis (‘output’) systems. Although our present activity involves only the eight languages of the present partners it is also conceived so as to facilitate the introduction of other member states' languages. If appropriate decisions are not made now, the languages of Europe will be catastrophically disadvantaged relative to English language forms, when the speech technology developments which are merely foreshadowed now become a reality in the next century. Standard European databases are already available on compact disk for five languages and a much larger data-gathering exercise has just been completed. Common workstation facilities using a standard configuration (SESAM) are being used in the laboratories of the project. These provide a reference for the application of a comprehensive range of software packages designed for input and output assessment in addition to the provision of supporting linguistic analysis software. Common methods of linguistic standardization have been evolved and a common computer compatible phonemic notation (SAMPA) has been introduced for European languages. A structured perspective for standards and resource-oriented themes in future speech and spoken language work is outlined.