Abstract:

Communication by voice over wired and radio systems is ever more important. Voice recognition by machine is commonplace today, whether it involves a person telephoning into an organization or the government spying on international communications. In these examples, context is not yet highly developed, so that the advantage a person has in understanding a sentence even if all words are not perceived does not ordinarily exist. Thus, correct understanding of isolated words is crucial. Two papers pioneered in establishing a metric for determining the degree of understanding of words in the presence of noise [French and Steinberg, ``Factors governing the intelligibility of speech sounds,'' J. Acoust. Soc. Am. 19, 90--119 (1947); Beranek, Proc. I.R.E., 35, 880--890 (1947)]. This paper will review the basic principles taught in those papers. Very low values of AI (i.e., little or no intelligibility) proved critical in studies on speech privacy in buildings by BBN scientists some 50 years ago. Those early studies also confirmed the relative importance of masking sounds in offices and other spaces, sometimes referred to as acoustic perfume, in achieving desired degrees of speech privacy.