Acoustic Description Research Articles

Baby talk: Acoustic description of infant's comfort-state vocalizations

Comfort-state vocalizations were recorded from 21 infants, seven at each of the ages of 3, 6, and 9 months. Spectrograms and digital methods were used to determine utterance durations, formant frequencies of vocalic segments, and characteristic features of phonatory behavior. Histograms of utterance durations show that the great majority of utterances had durations of 400 ms or less, although some utterances lasted a second or more. Volcalic utterances had formant frequency ranges of F1 = 0.6–1.4 kHz and F2 = 2.0–3.8 kHz for 3-month-olds, F1 = 0.5–1.7 kHz and F2 = 1.6–3.8 kHz for 6-month-olds, and F1 = 0.5–1.7 kHz and F2 = 1.4–4.1 kHz for 9-month-olds. Thus, the acoustic vowel space increased in both the F1 and F2 dimensions across the three age groups. Mean fundamental frequency was 445 Hz for 3-month-olds, 450 Hz for 6-month-olds, and 415 Hz for 9-month-olds. Laryngeal function varied substantially both within and across utterances, with some phonatory properties either differing greatly from adult speech or rarely if ever appearing in adult speech. Most infants had occasional vocal tremor, the frequency of which (9–21 Hz) was higher than that reported for adults (4–8 Hz). Infant vocalizations often were associated with rapid f0 changes, noise components particularly in the higher frequencies, alternating noise excitation, and biphonation (sudden appearance and disappearance of subharmonics, especially the 12 f0 series). Apparently, the infant drives the larynx to the extremes of its functional ranges, perhaps including nonlinear regions avoided by adults. [Work supported by NIH.]

Choices among traffic sounds with equal Leq levels: A replication

A previous pair of experiments produced a complex set of results which showed (a) that Leq, the energy‐equivalent sound level, predicts choices among traffic sounds better than do other traffic noise descriptors, and (b) time‐dependent sound‐level variability and average frequency spectrum shape are minor predictors of choice [G. W. Cermak and P. C. Cornilion, J. Acoust. Soc. Am. 59, 1412–1420 (1976); G. W. Cermak, J. Acoust. Soc. Am. 61, S56 (1977)]. The present experiments replicate both of the previous experiments using a new set of traffic sounds. In experiment I, people judged pairs of unconstrained traffic sounds; in experiment II, other subjects judged the same sounds which had been made equal in Leq level. Results supported the use of Leq as a traffic noise descriptor, but did not support the use of sound‐level variability or spectrum measures. Complex perceptual attributes of traffic sound with no simple acoustic description were the best predictors of choice when sounds were equal in Leq.

Preliminary analyses of speech sounds with a digital model of the ear

We believe that current lack of simple, invariant relations between the acoustic description of speech sounds and their phonetic features is due, in part, to the failure of current analyzers to adequately model the analysis performed by the auditory system. Our approach is to use models of the outer, middle, and inner ears for speech analysis. The sound-field-to-stapes velocity transformation is simulated with a digital filter consisting of two complex-conjugate pole pairs with center frequencies of 2.4 and 4.0 kHz and bandwidths of 0.7 and 2.0 kHz, respectively. This filter is consistent with descriptions of the outer ear provided by Shaw and of the middle ear by Zwislocki. The stapes is then used to drive a digital, 50-section implementation of the two-dimensional cochlear model developed by Spenner and Cox. Because the solutions of the model are in the time domain, its transient response is included in the speech-wave analysis and the need for arbitrary data windows and assumptions of periodicity are eliminated. Vowels and plosive consonants spoken by female and male talkers have been processed by the model. Sample traveling waves and their envelopes will be shown. [Supported by RR 00396 and NS 03856.]

Flapping rule in American English: an acoustic study

This study presents a detailed acoustic description of the /t, d/ flaps in American English, and the implications of this description for the formulation of phonological rules. The data base, a subset of the utterances described in the previous paper, contains minimal pairs of the types metal/medal, and polysyllabic words such as critical/medical. Our results show that dentals commonly labeled “flaps” fall into more than one category. For example, the durations and onset spectra of the /t/'s in waited and seating are measurably different, indicating a possible difference in the underlying production mechanisms. Our quantitative results suggest a reevaluation of the so-called flapping rule and its interaction with other phonological processes such as nasal deletion (as in winter) and vowel lengthing (as in waited/waded).

Time Domain Measurements for Speech Analysis and Recognition

At present, acoustical descriptions of speech sounds are usually given in terms of spectral characteristics, and spectral measurements are the usual form of data input for automatic speech recognizers. Measurements made directly on the acoustic signal are of interest as an alternate form of acoustical description and are potentially very useful in computer-based recognition schemes. A number of such measurements have been investigated. These include zero-crossing rate of the normal and differentiated acoustic signal, time interval between successive zero crossings for normal, low-passed, and high-passed versions of the acoustic signal, form factors of the waveform, and signal intensity Sound classes that have been studied include voiced and unvoiced fricatives, voiced and unvoiced stops, and vowels. Present results indicate that in comparison with spectral measurements: (a) rapid sequential features are better preserved; (b) the measurements are less sensitive to intensity fluctuations; (c) unvoiced fricatives and stops are easily characterized; (d) vowel resolution appears to be comparable. Application of these results to spoken word recognition is considered.

Speculations on performance models

The publication ofSyntactic Structuresin 1957 stimulated a much-needed re-evaluation among linguists as to the goals of linguistic theory and the nature of language. Part of the discussion which has ensued has centred around the question of linguistic competence versus performance. Competence has been related to performance as ‘langue’ is to ‘parole’. ‘Competence’ thus refers to the ‘underlying system of rules that has been mastered by the speaker-hearer’ (Chomsky, 1965) and ‘performance’ to the way the speaker-hearer utilizes this ‘internalized grammar’ when he actually produces and understands utterances. Despite the continued controversy about this distinction, little can be added to the justifications for it put forth over many decades (cf. Chomsky, 1957, 1964, 1965; Katz, 1964, 1966; Postal, 1966; Sapir, 1933; Levin, 1965; de Saussure, 1916; etc.). Yet there remains much vagueness as to the limits of each and the relationship between the two. For many years the confusion was due to the influence of Bloomfield who centred his attention on the speech act; his aim was the classification of the OUTPUT of performance, i.e. the utterances, and led to no theory about the dynamic process of performance itself (Bloomfield, 1924, 1926, 1927, 1933). While giving lip service to a concern for ‘langue’, his own mechanistic approach negated any possibility for the rules of ‘langue’ to be anything more than lists of recurrent patterns found in ‘parole’. And since he was of the opinion that ‘the physiologic and acoustic description of acts of speech belongs to other sciences than ours’ (Bloomfield, 1926: 153) he did not direct himself to those aspects of ‘parole’ which could explain speech performance.

Systems for compressing the bandwith of speech

Speech bandwidth compression is based on the present knowledge of human speech production and perception. This is the first in a series of papers reviewing the articulatory, acoustic, and network description of speech and discusses the bandpass compressor, which is a simple example of a bandwidth saving device.

Acoustical description of syllabic nuclei: an interpretation in terms of a dynamic model of articulation.

Measurements of formant frequencies throughout the vocalic portions of a number of phonemically symmetric consonant-vowel-consonant utterances have been performed using spectrum-matching techniques implemented on a digital computer. The contours representing the first two formants as a function of time are described in terms of several parameters, including initial and final frequencies, midpoint frequencies, durations, and measures of curvature. The data illustrate the manner in which each of these parameters is influenced by the features of the vowel and by place of articulation of the adjacent consonants. The results provide evidence that diffuse tense vowels in English are characterized by diphthongal asymmetric articulatory motions, whereas nondiffuse vowels are not, and that the influence of the consonantal environment on the articulation of the vowel is greater for lax vowels than for tense vowels. Some general statements are made concerning the extent to which contiguous vowels and consonants influence each other's production.

The Comparative Analysis of Sawari on the Shamisen

The shamisen is a long-necked, spiked lute from Japan. Like its Chinese ancestor, the san-hsien, the shamisen has three silk strings and a body covered with skin. Because the first Japanese shamisen players were biwa players, several important changes were made. Players began to use the plectrum used for the biwa instead of their fingers to pluck the instrument and found that the snake skin could not take the heavy blows of the plectrum, thus cat or dog skin was substituted for the snake skin (Malm 1959:185). Secondly, because the strings of the biwa produced a buzzing sound, this twangy quality was added to the shamisen.1 The innovations made by the biwa players serve to distinguish the shamisen from the san-hsien as well as provide the two most characteristic sounds associated with the shamisen; the percussive sound produced when the plectrum strikes the skin after plucking the string, and the buzzing quality known as sawari. Sawari might be described as a buzzing sound produced on the first string (lowest sounding string) of the shamisen. It is not only an inherent characteristic of the instrument, but functions as an important factor in determining the quality of the instrument. Musicians make value judgments on the shamisen by listening to the sawari it produces. The very fact that a shamisen produces this buzzing sound is not enough-an instrument may overemphasize sawari, or it may not produce enough. Yet, when asked to define the basis of their judgements, the musicians find it difficult to answer. It is the purpose of this paper to attempt to define sawari; to give its physical description and its acoustical description, and to determine the comparative properties within the sawari upon which musicians seem to base their judgments in determining the quality of a shamisen. The production of sawari is accomplished by two steps: first, the first string is removed from the top bridge, kamikoma (see Figs. 2 and 3 for the names of the parts), so that only strings two and three rest on the bridge and are free to vibrate (Fig. 4); secondly, the area just below the peg box, aze, is carved out slightly to form an indented area, tani, and a ridge, sawari no yama. The first string, being lower than the other two strings, rests lightly on the sawari no yama (Figs. 4 and 5). This interference causes the complex vibration which is termed sawari. There are three vibrating lengths of string which contribute to sawari (Figure 1):

An Acoustic Description of Dysarthric Speech

The paper reports the results of a spectrographic study of the speech of ten dysarthric subjects. The material recorded by each subject and analyzed spectrographically consists of a set of 50 CNC words, 60 words incorporating initial and final consonant clusters, and 50 spondees. The patterns associated with dysarthric speech were compared with spectrograms made of the speech of a normal speaker who pronounced the same set of test words, as well as with a considerable number of spectrograms of utterances by various other normal speakers of the same dialect, recorded and analyzed in the course of research in acoustic phonetics at the Communication Sciences Laboratory of The University of Michigan. Acoustic patterns characteristic of the speech of the individual dysarthric subjects are described; general symptoms associated with the speech of all subjects are presented and discussed. (This research was supported by a grant from the National Institutes of Health.)