Vocal synthesis in a computerized dictation exercise

Dictation can be considered as a transcription exercise of an utterance. The main difficulty for the elaboration of a dictation system consists in modeling the errors and the associated explanations provided to the learner. On these bases, an experimental DICTOR system is being developed as an assistant tool to learn the French language spelling. DICTOR includes an automatic checking tool based on a stochastic alignment algorithm and French written linguistic knowledge. This paper focus on both aspects. We discuss the spelling learning issues due to the no one-to-one correspondence between the utterance and the written text. Then the principles of the spelling difficulty groups (sdg) and associated explanation rules are presented. Finally, we report an observation of the DICTOR system in a school environment for the French language spelling. We then analyze how the different vocal synthesis influence the exercise and are perceived by the pupils in this new generation of language learning systems.