Text-to-speech synthesis (TTS) technology has numerous uses in voice-based information retrieval, desktop computer interfaces, language training, and a host of other applications. However, TTS systems are not yet used in many appropriate applications because their output quality is not acceptable to the average user. Funding for investigation of speech synthesis techniques in the United States has dwindled over the past several years. This has led to a slowing of progress in synthesis, and a a limiting of the scope of all speech technology education. This work will impact the broader scope of interactive systems research by (i) developing new speech synthesis techniques that allow more natural-sounding speech synthesis, and (ii) supporting the training of new investigators with a unique mix of expertise in signal processing, statistical modeling, and linguistic sciences. Course curricula, exercises, and freely distributed software tools will be developed to aid in this educational innovation. Research on data-driven signal processing techniques will strive to improve the quality of current TTS algorithms. The results of this research will be evaluated at multiple levels, and will impact interactive systems research and education at multiple sites throughout the world, aided by their distribution in the CSLU Toolkit.
