The following HTML text is provided to enhance online
readability. Many aspects of typography translate only awkwardly to HTML.
Please use the page image
as the authoritative form to ensure accuracy.
human subjects can then be used to provide validation at strategically chosen points.
Researchers at NTT and ATR in Japan have been especially prominent in these explorations, and their initial results look very promising. As such methods gain wider application, and especially as we see general availability of the large-scale single-speaker databases that will be required to support them, we can hope to see an increased rate of improvement in segmental speech synthesis quality. Thus, increased investment in speech synthesis research is warranted, both because there is an opportunity created by advances in microelectronics and because there are significant new ideas and new methods waiting to be applied.
As this research goes forward, it faces some pointed questions. What will it take to make synthetic speech that sounds entirely natural, or at least better than word concatenation voice response systems for restricted phrase types such as name and address sequences? Will progress come by a scientific route, through better modeling of human speech production, or by an engineering route, through larger inventories of prerecorded elements with optimal automatic selection and combination methods? How far can we push current ideas about text analysis algorithms? How can we produce more natural-sounding modulation of pitch, amplitude, and timing, and how important are such prosodic improvements relative to segmental improvements?
What will it take to put speech synthesis into true mass market applications? What will those applications be? Will the key development be cheaper hardware, a particular "killer" application, or better-quality synthesis? Will there be a gradual spread of the existing niche markets or a single breakthough?
How should we quantify progress in synthesis quality? What is the proper place for subjective testing relative to objective distortion metrics?
The papers by Carlson and Allen in this volume present a solid foundation of fact for evaluating these questions, and a wide variety of opinions were aired in the symposium discussion, from which an individual point of view has been distilled in this introduction. The next decade will be a lively and interesting time in the field of speech synthesis research, and there is little doubt that the situation will look very different 10 years from now.