Search

Attempts to mechanize character reading and speech recognition have greatly accelerated in the past decade. This increased interest was prompted by the promise of computer inputs more flexible in format than punched cards or magnetic tape. Research has shown that automatic sensing can be done reliably if the task is suitably delimited. Cleverly designed marks on standard forms can be both machine and man readable. A single type font or a few fixed ones are tractable if the print quality is controlled. Handprinting can be handled for careful writers, as can meticulous handwriting. Isolated spoken words taken from a small number of talkers and a limited vocabulary can be automatically recognized. Typical error rates for these machine-sensings run between 0.5 and 25 per cent. These results imply that reading unrestricted typestyles, handwritten scrawl, or recognizing conversational speech is beyond the reach of present methods. From the engineering viewpoint, questions of values enter. Might it not be wiser to punch cards or tape while making copy rather than depend upon complex character recognition hardware? Is it useful to have voice input to a computer when a finger and typewriter are available? Answers to such questions will depend upon the specific application. Certainly, the utility of automatic sensing will depend upon what is to be done with the material after it enters the computer as well as the internal organization of the machine itself.

This paper describes an implementation strategy for a vocal data management system (VDMS) being developed by the voice input/output project at the System Development Corporation. VDMS will accept connected speech of a language describable by 25-50 phrase equations and having a vocabulary of approximately 1000 words formed from about 100 data records. The strategy is based on the concept of predictive linguistic constraints (PLC). The present concepts of fixed directionality in parsing are replaced by a more generalized approach. To facilitate this flexibility, the system comprises a set of near-independent coroutines that are interconnected by a software busing structure. The VDMS acoustic processors verify the predictions. Very loose matching criteria are used for locating the predicted words. Special attention is given to word segments that are experimentally determined to be most invariant.

An on-line digital computer processing system is considered in which an ordinary telephone is the complete terminal device, input to the computer being provided as a sequence of spoken words, and output to the user being audio responses from the machine. The feasibility of implementing such a system with a FORTRAN-like algebraic compiler as the object processor is considered. Details of a specific word recognition program are given. This technique depends on three simplifying restrictions, namely, a “small” vocabulary set, “known” speakers, and a “moment of silence” between each input word. Experimental results are presented giving error rates for different experimental conditions as well as the machine resources required to accommodate several users at a time. The results show that at this time it is both economically and logically feasible to handle at least 40 users at a time with an IBM 360/65 computer.

O Dopping London: OUP 1971 pp 399 price £4.25 This is an extremely good book – so much so that the title is an understatement of its contents. A better title would have been 'A complete course in computer appreciation', since the book covers the whole field comprehensively from early card sorting machines to the latest techniques of character processing, visual display and voice input/output.