Voice Communication Between Humans and Machines - Science and Prescience

Humans depend upon information. As the complexity of society increases, so does the human need to sort, select, assimilate, and respond to information. Computers can aid humans - especially if computational power is matched to sensory capabilities. Speech and hearing are key senses for human information exchange. And, machines that match these faculties can aid information flow. Machines that talk and listen - and can 'understand' to the extent of carrying on limited (but intelligent) conversation - are rapidly evolving. Advances are being fueled by new understanding in analysis, coding, transmission and display of information, in language modeling and speech acoustics, and by the ever-increasing capabilities and economies in computation. The coming decade will consequently see high-quality text-to-speech conversion for multiple languages, speakers and dialects. Automatic speech recognition will permit fluent natural-language conversion with machines on delimited topics. Digital coding, of high efficiency and low error vulnerability, will support new services in electronic voice mail, as well as pervasive deployment of cellular radio. And, sophisticated sensors, such as auto-directive microphone systems, will make high-fidelity speech capture in noisy reverberant environments less problematic. But, the totally-fluent intelligent computer of 2001, conversing on all topics with all persons, will not yet be realized, as neither will flawless automatic translation of spoken language. To move technology towards this sophistication requires visionary choices in research directions today. This discussion draws a perspective on evolving capabilities in human/machine communication - and on the research directions that will help us invent our future.