Multidimensional Model for Speech Recognition
B. V. Bhimani · The Journal of the Acoustical Society of America · 1963
Speech sounds do not appear randomly in utterances. Rather, there appears to be a set of preferred positions for these sounds. They also have a set of determinable acoustic modifications that these sounds receive from their environment and impose upon it. That is, as a language consists not only of a vocabulary but also of a grammar that limits its possible verbal combinations, so does speech consist not merely of a set of sounds but also of an embedded or intrinsic structure within which these sounds form sequences. Such a structure is partly the result of capacities and limitations of the speech-producing mechanisms and it is partly shaped by linguistic tradition. It seems possible, therefore, to describe speech acoustic phenomena by a set of logical formulations that form the analytic correlative of the intrinsic sound structure. Moreover, such formulations should be subdivisible in terms of the degrees of freedom of speech-producing organs. Such all organization of rules is an orderly integration of data on the genetive, phonetic, phonemic, and acoustical aspects of speech, and it constitutes a multidimensional model for speech that combines the coherence of analytic form with fidelity to the realities of speech events. [This work was supported by Air Force Cambridge Research Laboratories, U. S. Office of Aerospace Research, under contract No. AF 19(628)-2766.]