Digital machines-computers-offer many advantages in communications and information processing. They can accomplish computations at high speed, and they can store and rapidly access vast amounts of information. Despite this sophistication, they are not adept at communicating with humans in a way humans find convenient-namely, by natural spoken language. Computers traditionally prefer the special symbols of compilers and assemblers typically communicated via a typewriter keyboard and a printed display. But, if they could be given human-like abilities for voice communication, their value and ease of use for humans would increase. And, the ubiquitous telephone would take on more of the capabilities of a computer terminal. Giving the computer a "mouth" to talk t o humans draws upon the techniques of speech synthesis. Giving the computer "ears" to listen to humanspoken commands entails automatic recognition of speech. T o provide computers the ability for natural language exchanges, the engineer can be guided to a large extent by the processes of human speech generation and recognition. The physics of sound generation in the human vocal tract and the signal analysis performed in the peripheral auditory mechanism are relatively well understood and can be used as design guides for synthesizers and recognizers. By contrast, the more central cerebral functions of message formulation and message comprehension are less well known, and this lack is reflected in the limitations that presently characterize machine synthesis and recognition. This talk assesses the progress of speech synthesis and recognition by computer techniques, and it outlines potential applications to new voice-communication services. Finally, it draws a perspective on foreseeable developments, and on the continuing research that will support increased sophistication in human/machine communications. At last year's symposium a number of groups, including RSRE, reported the development of monolithic SAW convolvers with attractive IEEE TRANSACTIONS ON SONICS AND ULTRASONICS, VOL. SU-29, NO. 3, MAY 1982 159 performance characteristics: integration time -10 ps, bandwidth -100
SAW ELASTIC CONVOLVERSChairpersonMHz, efficiency --75 dBm. In the intervening period we have introduced a number of refinements to the design which have improved the performance in all these primary respects, and in many secondary aspects too. At the time of writing we have produced a convolver with integration time cxcceding 15 ps, and with efficiency exceeding -70 dBm over a bandwidth of 120 MHz. These improvements will be described, together with improvements to many secondary aspects such as, reduced size of the lithium niobate "chip" and its package, simpler testing and setting-up, improved suppression of unwanted selfconvolution signals, and the important aspect of improved reproducibility ofthe overali device ckclracrerisrics. It is now well established that the elastic convolver is a simple, compact, and inexpensive structure which promises to perform, with realtime programmab...