Speech Algorithms for Clinical and Educational Applications

The main goals of the proposed research are to improve speech synthesis technologies so as to make them suitable as supporting technologies for educational and clinical applications, and to demonstrate the benefits of speech technology based tools for these applications.
In many of the envisaged applications, synthetic speech feedback is important in that it frees the pupil’s gaze to scrutinize the lesson’s text or graphical material. Moreover, an important advantage of systems, like a computerized reading tutor is its ability to provide immediate auditory feedback. An important demand for the generated speech in general and for language learning and reading tutor applications in particular, is that it sounds natural and highly intelligible. In order to meet the quality standard while avoiding the need for developing a complete text-to-speech synthesizer, a part of the research will be devoted to the design of a hierarchical speech generation strategy which will compose speech as a sequence of units of different sizes (diphones, syllables, words, phrases, sentences). In the case of corrective feedback, the generated speech must over-emphasize certain sounds that were pronounced incorrectly by the user. For that reason, we will search for appropriate speech modification strategies that are not available in conventional speech generation systems.