From a practical point of view, automatic speech recognition involves the speaker at least as much, if not more, than it involves the mechanism. I will try to illustrate this point by describing some of the strategies which have evolved during work on a word-recognition computer program. Many of the details that I may omit are included in my report for the Research Laboratory of Electronics of M.I.T. -Word-Recognition Computer Program.- Before getting into the specifics of the work, I would like to say a few words about the overall aims of such work. In my own experience, there are three areas for which automatic speech recognition, and, in particular, automatic word recognition, may be pertinent. These are, speech bandwidth compression, voice-actuated mechanisms and -talking to computers.- I believe that the last of these three possible applications is presently the most practical and pertinent one. Most of my recent efforts have been directed towards supplying a useful verbal command program as an aid to a graphical communications system whereby the computer operator can draw flow charts on the oscilloscope with either a light pen or a Rand tablet and have the resulting drawing translated into a program by the master program. Such an aid appears to be very useful for mode changes in the master program. Chronologically then, the work proceeded as follows: first, the development of a set of rules to be used as the basis of a word-recognition scheme, second, the development of a strategy which appeared to be appropriate for on-line control of other programs. Finally, and this is the step we have not yet taken, is the actual connecting of the word-recognizer to the graphical communications program.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.