Towards Establishing a Methodology for Benchmarking Speech Synthesis for Computer-Assisted Language Learning (CALL)

1. Introduction In very simple terms, speech synthesis is the process of making the computer talk. As such, speech synthesisers offer another means of providing spoken language input to the learner in CALL environments. Indeed, many potential benefits (ease of creation and editing of speech examples, generation of various kinds of modified input, generation of speech models and feedback on demand, etc.) and uses (as a reading machine, as a model which the learner can use to familiarise him/herself with segmental and suprasegmental distinctions in the Target Language (TL), a model to imitate in pronunciation training and reading exercises, feedback in pronunciation training and reading, and as the voice of a conversational partner in communication-based activities) (Hamel and Handley, to appear) of speech synthesis in CALL have been put forward. Yet, the use of speech synthesis in CALL is not widely accepted (Sobkowiak, 1998; Egan and LaRocca, 2000), and only a few applications have found their way onto the market (for example, the Etaco hand-held talking dictionaries, talking dictionaries for PC including Oxford-Hachette French dictionary on CD-ROM, and The Longman Dictionary of American English for Macintosh, and the CALLware Spanish for Business Professionals, etc.). In my opinion, people are right to be sceptical. The use of speech synthesis in CALL is not proven. Very few evaluations of speech synthesis for and in CALL have been conducted. One potential reason for the neglect of evaluation is the fact that it competes for time and resources which could be spent on further research and development (Hirschman and Thompson, 1996). In order to achieve a balance between development and evaluation,