Man-machine communication by voice is no longer a far-fetched notion Indeed the ' talking and listening' machine has already left the frontiers of science fiction fantasv. This paper outlines the limitations of existing means of commumcation with computers and the background to developments in voice input/output technology. Within the area of man-machine communication by voice there are three main avenues of research: speech synthesis (voice output), speaker recognition (identification and verification), and speech recognition (voice input). The technology common to these developments is speech encoding-digital representation of voice signals for processing by a computer. Techmques of speech encoding and speech synthesis are relatively well developed Speech recognition technology, however, is still in its infancy The recogrution task is complicated chiefly by the fact that characteristics of the speech signal differ from person to person, and even in the speech of one individual on different occasions. Because of this, it is more practical to divide speech recognition into four smaller categories, according to whether the systems are speaker-dependent or independent and recognized isolated words or continuous speech. Most of the commercially available systems are speaker-dependent/Isolated word systems. Up to now such systems have been hamstrung by poor recognition accuracy, insufficient vocabulary size-typically in the region of 100 to 200 words-and high implementation cost. Fortunatelv, the situation is improving rapidly and, indeed, claims for recently marketed systems sound most impressive-accurate recognition of up to 1000 words or more. The ultimate goal in speech recognition is speaker-independent/continuous speech systems which, although technologically feasible at the present time, may hot be available commercially for some time yet. And finally, speaker recogmtion can be regarded as a special case of speech recognition: nevertheless it is different in terms of its practical applications. . Existing products are particularly swtable for ' hands-busy. eyes-busy' operations and indeed they are used in several manufacturing applications and also as aids to the disabled They have also made a significant impact in voice store-andforward applications. Those who are planning to use Voice I/O systems have a wide range of technological options from which to choose These include complete turn-key systems, penpheral attachments to one's own computer, plug-in boards, chips and even software packages. However most potential users are taking a cautious approach because of technological and economic uncertainties in the speech product market. There is no reliable guide for users to assess the standards of performance of individual systems. There are also several other problems still to be resolved before speech technology products become widely applicable. These range from computation and technical problems-phonetic, semantic and syntactical-to user acceptance/interface of the end-products.Part II (following paper in t...

Following the description in Part I of the major developments in voice I/O technology and of the problems still to be overcome, Part II discusses the practical applications of the technology, both actual and potential. Interest in speech technology is international, and applications have been found worldwide, in many situations. These range from a &dquo;real time telephone translation system&dquo; demonstrated in Japan, which &dquo;will turn Japanese into English, French and Korean&dquo; to a train time-table information service accessible by telephone, developed by British Telecom. At present, most voice recognition systems are at work in industry, as they are ideal for situations in which both hands and eyes are busy and not available to wnte down information or operate a keyboard. In the office the main impact will be in the area of voice mail, permitting the elimination of ' telephone tag' and an increase in overall efficiency of communication. Word processing functions can also be aided by the addition of voice recognition, particularly in the editing of correspondence. Indeed, speech controlled typing systems ('automated secretary') are expected before the end of the decade. Computers that talk and listen' are also an advantage to the disabled: systems are being designed to cater for their special needs, and can be of value to the blind and heanng-impalred as well as to those severely physically handicapped Speech technology applications exist equally in medicine and surgery, and also in aviation and the space industry.Finally, in library and information services, direct voice input/output of data can allow radical changes in any or all areas currently employing computers, such as technical processes, ordering and receipt of material and stock control, but the most far-reaching implications are for on-line information retrieval services The most immediate and promismg development will be the use of voice I/O interfaces in online searchmg which, apart from other benefits, will allow flexibility of command to the user. Reference has already been made to systems allowing direct access by telephone to remote databases Should these eventually become available for use by any individual, not only online search services, but all information provision whether in libraries or information services, could be radically and irrevocably altered.

No abstract