Invented spelling and speech synthesis feedback in a deep orthography - Considerations in designing a speech synthesis feedback.

Purpose: Intervention studies suggest that beginners using invented spellings advance more rapidly if they receive corrective feedback (e.g. Ouellette & Sénéchal, 2008). A speech synthesis software can read aloud children’s invented spellings and thereby provide immediate feedback, but designing software that can supplement or even replace teacher feedback, involves non-trivial decisions. The present paper discusses issues in the design of synthetic speech for Danish beginners e.g. which phones to assign to letters, since there isn’t a simple one to one relationship between letter and phonemes in Danish orthography. Method: A list of required characteristics of the synthesis was made. The degree to which these requirements was meet by two major types of synthesis (Analytic vs. Unit selection) was evaluated, and strengths and limitations were weighted against each other. Results: An analytic synthesis was preferred, since this gave access to the phonetic level of the synthesis. This made it possible to ensure that the synthesis assigns the phone that is known from the letter name to each letter. At the same time, the synthesis was allowed to violate this principle, whenever it encountered a real word. The synthesis read at a slow rate, which resulted in an audible blend of individual phones. Conclusions: It was possible to develop a synthesis, with characteristics that fulfill the theoretically based requirements. The results from an ongoing intervention study will reveal whether the feedback provided by the synthesis is more effective than traditional teacher feedback in stimulating kindergarten children’s early literacy skills.