Gordon E. Peterson

Automatic Speech Recognition Procedures

This paper is concerned with the transformation of the varying acoustical parameters of speech to a discrete code to form the printed output of an automatic speech recognizer. The development of general automatic speech recognition procedures requires a definition of the linguistic code to be transcribed, and a statement of the dialectal and other conditions under which the recognition is to be achieved. Essential procedures in automatic speech recognition include: the analysis of the input speech wave into a series of basic acoustical parameters in frequency; the representation of the normalized parameters by a set of phoneme and prosodeme candidates by reference to stored linguistic information; and the print-out into words separated by spaces and grouped by means of a set of punctuation marks. The possibility is considered of employing values of conventional spelling and punctuation in the automatic representation of spoken American English.

Objectives and Techniques of Speech Synthesis

The synthesis of speech is discussed as one of the simpler problems of language automation While ultimately speech synthesizers will doubtless have many practical applications, their chief value at present is in basic research on the relation of speech parameters to linguistic judgments. Two basic methods of speech synthesis are considered: 1) the generation of speech from stored segments, and 2) the generation of speech through continuous control of the various speech parameters individually; in the latter case, the parameters may be physiological or acoustical. It is concluded that electronic analogues of the physiological speech mechanism provide a means of evaluating hypotheses about the physiologic – acoustic speech transformation, and that acoustical speech simulators are the most realistic and practical research tools for the experimental study of speech perception.

Recurrently Impulsed Resonators in Speech and Psychophysical Studies

The need for data on the perception of sounds produced by the excitation of resonance systems is discussed in relation to speech and psycho-acoustics. The voltage response of a series of N decoupled low-pass resonator sections to the sudden application of various types of input pulse trains is reviewed. The output of such a series of resonators to a recurrent impulse which is suddenly applied contains a d.c. term, a series of transient terms at the uncoupled resonator frequencies, and a steady-state term involving the harmonics of the input pulse train. A circuit for psychophysical tests is described which provides a number of discrete positions of resonator frequency and damping. Sound spectrographic analyses demonstrating the transient and steady-state terms for a single resonator are presented.

Intrinsic Duration of Syllable Nuclei in English

The study deals with the influence of preceding and following consonants on the duration of vowels and diphthongs of American English that may occur in stressed position within a sentence. A set of 1263 CNC words, pronounced in an identical frame by the same speaker, has been analyzed spectrographically, and the influence of various classes of consonants on the duration of the preceding nucleus has been determined. The residual durational differences are analyzed as intrinsic durational characteristics, associated with each syllable nucleus. The theory is tested with a set of 30 minimal pairs of CNC words, uttered by five different speakers.

Diphthongs, Glides, and Transitions

The study deals with the formant movements associated with diphthongs, glides, and consonant transitions in spoken American English. The transitional characteristics associated with all initial and final consonant phonemes have been studied in a large sample of utterances. The rate of change of the transition from a consonant hub to the steady state vowel formant position and conversely has been investigated for vowels which are commonly considered monophthongs. It is assumed that such changes are cues for the perception of the consonants rather than linguistically significant components of the vocalic nucleus. The rate of change associated with syllable nuclei commonly perceived as diphthongs ([ɑɪ, ɔɪ, ɑᴜ]) is investigated and compared with the changes due to consonant transitions in monophthongs. An attenpt is made to establish criteria by which the formant movements due to transitions may be distinguished from movements that have linguistic signaling value within the syllable nucleus.