Search

Nonlinguistic information about the speaker's emotional state is conveyed in spoken utterances by means of several acoustic characteristics and listeners can often reliably identify such information. In this study we investigated the effect of short- and long-term F0 measures on perceived emotional stress using stimuli synthesized with the LPC coefficients of a steady vowel and varying F0 tracks. The original F0 tracks were taken from naturally occurring speech in highly stressful (contingent on terror) and nonstressful conditions. Stimuli with more jitter were rated as sounding more hoarse but not more stressed, i.e., a demonstrably perceptible amount of jitter did not seem to play a role in perceived emotional stress. Reversing the temporal pattern of F0 did not affect the stress ratings, suggesting that the directionality of variations in F0 does not convey emotional stress information. Mean and maximum F0 within an utterance correlated highly with stress ratings, but the range of F0 did not correlate significantly with the stress ratings, especially after the effect of maximum F0 was removed in stepwise regression. It is concluded that the range of F0 per se does not contribute to the perception of emotional stress, whereas maximum F0 constitutes the primary indicator. The observed effects held across several voices that were found to sound natural (three male voices and one of two female ones). An effect of the formant frequencies was also observed in the stimuli with the lowest F0; it is hypothesized that formant frequency structure dominated the F0 effect in the one voice that gave discrepant results.

The Origins of Complex Language: An Inquiry into the Evolutionary Beginnings of Sentences, Syllables, and Truth. Andrew Carstairs-McCarthy. Oxford. Oxford University Press, 1999. 260 pp.

Studies of human development and swallowing show that a supralaryngeal vocal tract (SVT) capable of producing quantal vowels involves (1) facial restructuring yielding a short oral cavity; (2) a tongue that moves down into the pharynx carrying the larynx down with it; and (3) a long neck. The skeletal features of hominid fossils suggest the absence of fully human speech anatomy until 100 000 years ago. With long faces and short necks, Neanderthals and early anatomically modern humans could not have possessed SVTs capable of swallowing and of producing fully human speech. Probable [i] SVT shapes were computer‐modeled from perturbations of vocal‐tract area functions obtained from MRIs of adult humans. When normalized to the length of an adult human SVT, F1/F2 patterns fall outside of the range of Peterson and Barney’s [‘‘Control methods used in a study of the vowels,’’ J. Acoust. Soc. Am. 30, 739–742 (1952)] data used by Lieberman and Crelin, ‘‘On the speech of Neanderthal man,’’ Linquistic Inquiry 2, 20...

The visual abstraction procedure used in previous studies of declination was tested using 12 subjects who each fit the F0 contours of 19 spoken short simple sentences with baselines. These baselines were found to be poorly replicated by the fitters. An objective all‐points least‐squares best‐fit procedure was tested on this corpus and on a set of sentences that had been produced in both spontaneous and read speech by six speakers. The all‐points linear regression line was a better descriptor of the F0 contours than either baselines or toplines. Declination did not always occur in these simple declarative sentences; there was more variation present in the F0 contours of sentences that had been uttered during spontaneous speech; 35% of the spontaneous sentences did not show declination; 45% of these sentences better fit the breath‐group model. Their F0 contours could be described by a level all‐points linear regression line followed by a falling terminal segment.

ed from them regarding the ad hoc reconstructive process in general, we simply doubt the validity of Lieberman's conclusions [1978:368]. The reader should know that a detailed discussion of the basis for the reconstruction of the Neanderthal supralaryngeal vocal tract is presented on pages 124-139 of Lieberman (1975). Carlisle and Siegel are again incorrect in their reference to a prefatory comment on page 123, which serves as an introduction to the detailed discussion, as though it were the whole discussion: ". .. we merely have the statement that the reconstructions were made simply on the basis of. . . the homologies that exist between. .. [the reconstructed skulls] and those of living forms." Carlisle and Siegel's confusion concerning the anatomical material on which the La Chapelle-aux-Saints vocal tract reconstruction was based on should also not be transferred to the reader. The reconstruction

In claiming that biology is not likely to be a branch of the present physics, I do not wish to imply that life can in some mysterious way evade the laws of physics. I believe the situation is comparable to the problem of electricity and magnetism as it appeared before, and even during, the time of Maxwell. . . . He understood only later that electric and magnetic fields were important and basic concepts in physics, not contained in the concepts of mechanics, but of course not in contradiction to the laws of mechanics-and that

The rate of change of fundamental frequency with respect to transglottal air pressure was determined for a single male speaker by sinusoidally varying buccal air pressure while the speaker sustained short episodes of pronation at various fundamental frequencies. The speaker phonated at two levels of “effort,” i.e., “soft” and “loud” phonation with and without auditory feedback. The sensitivity of the larynx to variations in transglottal air pressure varied from 3 to 18 Hz/cm H2O. Fundamental frequency was most sensitive to variations in transglottal air pressure at high frequencies and in the “soft” mode of phonation. The minimum transglottal air pressure for sustained phonation was 2–3 cm H2O. These results are consistent with some earlier studies and a recent theoretical model of laryngeal activity. These results further indicate that variations in subglottal air pressure and adjustments in laryngeal muscular tension both play a role in regulating fundamental frequency during normal speech.

Some representative vocalizations of captive rhesus monkey, chimpanzee, and gorilla were recorded and analyzed by means of sound spectrograms and oscillograms. It was found that these animals' vocal mechanisms do not appear capable of producing human speech. The laryngeal output was breathy and irregular. A uniform cross section, schwalike configuration appeared to underlie all the vocalizations. These animals did not modify the shape of their supralaryngeal vocal tracts by means of tongue maneuvers during a vocalization. Formant transitions occurred in some vocalizations, but they appeared to have been generated by means of laryngeal and possibly velar or lip movements. The nonhuman primates lack a pharyngeal region like man's, where the cross‐sectional area continually changes during speech. The data suggest that speech cannot be viewed as an overlaid function that makes use of a vocal tract that has evolved solely for respiratory and deglutitious purposes; the skeletal evidence of human evolution shows...

The sounds of human speech make human language a rapid medium of communication through a process of speech “encoding.” The presence of sounds like the vowels [a], [i], and [u] makes this process possible. The supralaryngeal vocal tracts of newborn Homo sapiens and chimpanzee are similar and resemble the reconstructed vocal tract of the fossil La Chapelle-aux-Saints Neanderthal man. Vocal tract area functions that were directed toward making best possible approximations to the human vowels [a], [i], and [u], as well as certain consonantal configurations, were modeled by means of a computer program. The lack of these vowels in the phonetic repertories of these creatures, who lack a supralaryngeal pharyngeal region like that of adult Homo sapiens, may be concomitant with the absence of speech encoding and a consequently linguistic ability inferior to modern man.

Pitch perturbations were computed by measuring the differences between the durations of adjacent fundamental periods from recorded acoustic waveforms. Sound‐synchronized high‐speed motion pictures of the vocal cords were also taken with a laryngeal mirror. It was found that pitch perturbations reflect variations in the waveshape of the glottal area wave, as well as variations in glottal periodicity. Perturbations having an absolute value ⩾0.5 msec are apparently induced by transients in the air‐pressure drop across the glottis, which may in turn be caused by changes in vocal‐tract configurations. Speakers with longer fundamental periods tend to have larger perturbations. The pitch perturbations of 23 speakers with pathologic larynges were measured. Certain of the speakers who had pathologic growths on their vocal cords had larger pitch perturbations than did normal speakers with the same median fundamental periods. The “perturbation factor,” which was defined as the percent that perturbations ⩾0.5 msec oc...

Pitch pulses were electronically derived from the utterances of three male native speakers of American English who each read eight neutral test sentences in certain “emotional” modes, i.e., as a question, an objective statement, a fearful utterance, a happy utterance, etc. A fixed‐vowel POVO‐type synthesizer was excited by these pitch pulses. The pitch perturbations, or rapid variations in the fundamental excitation rate, could be smoothed out and the POVO could be amplitude‐modulated with a signal derived from the original speech envelope amplitude.Tapes were recorded and presented to separate groups of naive listeners who categorized the emotional modes in forced judgment tests. Results of the tests show that with unprocessed speech, the listeners were able to correctly identify the emotional content 85% of the time. When only pitch information was presented, correct identification was made 44% of the time. When amplitude information was added to the pitch information, the identification rose to 47%. Sm...

The acoustic theory of speech production relates articulatory maneuvers to acoustic attributes of speech. Some procedures are discussed that make use of this theory to analyze the vocalizations of nonhuman primates. These procedures make use of sound spectrograms, oscillograms, and computer-implemented analogs of the primate vocal apparatus as well as anatomical measurements. The use of these techniques in recent studies of nonhuman primate vocalizations is reviewed. These studies show that nonhuman primates lack the anatomical apparatus that is necessary for the production of the full range of human speech. Some unresolved questions concerning the structure of nonhuman primate utterances are discussed with regard to human linguistic ability.

The paper that ’t Hart disputes [Lieberman et al., J. Acoust. Soc. Am. 77, 649–657 (1985)] refuted the general claim of declination theory, i.e., that the fundamental frequency always falls throughout an utterance. We are pleased to see that our paper has had the appropriate effect and that ’t Hart and other proponents of the declination theory have modified their stand and no longer hold to this extreme position. We did not claim that a lowering of fundamental frequency ‘‘only occurs in the terminal part of the utterance.’’ ’t Hart’s comments obscure the other points that we demonstrated in our paper. We showed that hand‐drawn declination lines placed on F0 contours have different slopes when different observers draw them. To paraphrase a relevant line from Alice in Wonderland, hand‐drawn declination lines can show anything you want them to show. The auditory matching procedure that ’t Hart uses in his own studies introduces additional possible errors that make his analyses almost meaningless.

vocalizations of chimpanzees (Lieberman, Crelin, and Klatt 1972:298-299) which were analyzed in detail in two of our previous papers (Lieberman 1968; Lieberman et al. 1969). These animals, in fact, make use of a subset of the phonetic contrasts that our computer simulation predicts. The computer simulation is thus, if anything, too generous. We discuss (Lieberman et al. 1972:297-298) the actual vocalizations of newborn humans which are again analyzed in detail in Lieberman, Harris, et al. (1972), as well as by Irwin (1957). The results of the simulation match with the actual vocalizations of newborn humans. We did not dwell on the

The set of nine nondiphthong vowels of American English were synthesized using the averaged male values from Peterson and Barney (1952). These vowels were produced in three conditions of the fundamental frequency: (1) average 135 Hz, (2) low 100 Hz, and (3) high 250 Hz. In forced‐choice testing, subjects identified vowels in the average and low condition of the f0 with greater accuracy than the high f0 vowels. A second experiment was conducted using the female formant frequency values from Peterson and Barney and the same conditions of f0. Subjects still performed better on the low and average condition of the f0 for these vowels. These results suggest that the human formant frequency extractor is aided by the more detailed transfer function of a lower fundamental. The pattern of vowel errors was not the same across f0 conditions nor was it the same across male or female formant values. The point vowels [i], [u], and to a lesser degree [a] were the most consistently identified across conditions of the fundamental.

In recent work Blumstein and Stevens [J. Acoust. Soc. Am. 66, 1001–1017 (1979); 67, 648–662 (1980)] have proposed that the short‐term onset spectra at stop‐consonant releases provide invariant cues for the perception of place of articulation, and that the salient acoustic characteristics of these cues can be characterized by a set of templates. The present study investigated the suitability of these templates for characterizing place of articulation in utterance‐initial voiced stop consonants produced by two children 71–75 weeks old in spontaneous discourse. The overall correct categorization rate for the templates was 65%, with 70% correct categorizations for the bilabial template, 85% for the alveolar template, and 40% for the velar template. The results of a perceptual study in which listeners were asked to identify place of articulation for computer‐edited 15‐, 25‐, 40‐, and 135‐ms samples of the initial parts of these utterances were compared with the results of the template matching study. Although listeners’ identification of place of articulation was at well above chance level on even the 15‐ms tokens, there was no correlation between the results of the template‐matching study and the perceptual data. The perceptual data thus indicate that short‐term onset cues exist in the spontaneous speech of children at an early stage of language development, however, these cues do not appear to be adequately characterized by the Blumstein–Stevens templates.

Subglottal air pressure during speech was measured by two methods: (1) by recording from a tracheal catheter, and (2) by recording from a 10‐cm‐long balloon positioned in the middle third of the esophagus and correcting for the pressure developed by the elastic recoil of the lungs. A simple linear elastic‐recoil correction factor was devised. The two measures of subglottal air pressure yielded similar results. Some of the conditions that can produce pressure artifacts in esophageal balloons are discussed. The data also show that peaks in subglottal air pressure are not always associated with phonetic prominence, particularly when several noncontiguous words are emphasized in the same breath group.

Mehler takes an extreme Chomskian position and disputes the biological premises of my book. The Biology and Evolution of Language (Harvard U.P. Cambridge, MA). These premises are consistent with modern biological thought and data. Mehler distorts the tone, content, and intent of my book.

The reader will readily note that much of Repp’s criticism [J. Acoust. Soc. Am. 78, 1114–1116 (1985)] reflects his contention that there has been a procedural transgression on the part of the authors and the Society that establishes a ‘‘dangerous precedent.’’ Since Repp provides a partial account we feel that it is first necessary to address this issue.

The theory proposed in Lieberman (1967, Intonation, Perception and Language, MIT Press, Cambridge, Mass.) states that both subglottal air pressure and laryngeal muscular maneuvers control fundamental frequency. Two phonetic features, the breath group and prominence, account for some of the linguistic functions of intonation. In the archetypal unmarked − breath group, fo follows air pressure. The archetypal articulatory correlate of the marked + breath group is an increase in laryngeal tension. Recent electromyographic data show that laryngeal muscles are responsible for the terminal fo rises of + breath groups and nonterminal + prominences in − breath groups. In contrast, fo peaks of + prominent syllables in + breath groups are caused by air pressure peaks. Independent data of Fromkin and Ohala (1968, Kyoto Speech Symp.) are consistent with these electromyographic observations. The Fromkin and Ohala data also show that the mean rate of change of fo with respect to air pressure is 12.5 Hz/cm H2O. These dat...

Synthetic three formant vowels were recorded with thirty‐two different intonation contours, by varying the fundamental frequency as a function of time. The perceptual effects of these variations were investigated in a test in which listeners were required to discriminate between different intonation patterns. The reliability of the discriminations was correlated with the direction in which the intonation contour shifted, the rate of the shift and its magnitude. The relevance of the results to the problem of English sentence intonation is briefly discussed. [This work was supported in part by the U. S. Army (Signal Corps), the U. S. Air Force (Office of Scientific Research, Air Research and Development Command), the U. S. Navy (Office of Naval Research), and the National Science Foundation.]

Comparative anatomic and acoustic studies of the supralaryngeal vocal tracts of living and extinct primates indicate that speech production has been an important factor in human evolution. The sounds of human speech make human language a rapid medium of communication through a process of speech “encoding.” The presence of the human vowels /a/, /i/, and /u/ facilitates this process. The supralaryngeal vocal tracts of newborn Homo sapiens and the living nonhuman primates are similar and resemble the reconstructed vocal tracts of fossil Australopithecine and Neanderthaloid hominids. Vocal tract area functions that were directed towards making the best possible approximations to the human vowels /a/, /i/, and /u/, as well as certain consonantal configurations, were modeled by means of a computer program. The lack of these vowels in the phonetic repertoires of these creatures, who lack a pharyngeal region like that of adult Homo sapiens, may be concomitant with the absence of speech encoding and a consequent l...

The activity of the larynx is a function of the tensions of the laryngeal muscular system, the mass of those parts of the vocal cords that are in motion, the losses caused by the collision of the vocal cord's margins, and the aerodynamic forces that act upon the vocal cords. The larynx is coupled through these aerodynamic forces to the rest of the respiratory system. The subglottal air pressure is affected by the resonances of the lungs and trachea. When the fundamental frequency of the glottal waveform or the lower harmonics of the fundamental frequency excite these resonances, the peak subglottal air pressure builds up. The superglottal air pressure is a function of the air flow through the vocal tract. Pressure transients result when constrictions occur during voicing. Variations in the glottal waveform, therefore, result from period to period, which causes rapid variations in the glottal source's spectrum and periodicity. This paper discusses the results of analyses of the acoustic waveform and high‐s...

High‐speed motion pictures of the larynx were exposed by the techniques of indirect laryngoscopy at approximately 10 000 frames/sec. The acoustic waveform at the speaker's lips was also optically recorded on the film. The area of the glottal opening and amplitude of the acoustic waveform were measured, with the aid of a digital computer, for each frame of a film in which the speaker uttered /ɑ/ for approximately 0.3 sec. This short‐voiced sound is about the duration of the voiced segments that normally occur in fluent speech. The changes of the glottal‐area waveform are correlated with the beginning and end of phonation and the positions of the arytenoid cartilages.

While researchers applying quantitative techniques to the characterization of intonation patterns [W. E. Cooper and J. M. Sorensen, Fundamental Frequency in Sentence Production (Springer‐Verlag, Berlin, 1981)] have claimed success in relating phenomena such as F0 “resetting” to the presence of syntactic boundaries [J. Breckenridge and M. Y. Liberman, “The declination effect in perception,” unpublished manuscript, Bell Laboratories (1977)], this procedure has been confined to a highly limited corpus of short, simple‐declarative read sentences. The degree to which these models can be generalized to spontaneous speech has recently become the subject of controversy [N. Umeda, J. Phonet. 10, 279–290 (1982)]. We therefore examined a corpus of spontaneous and read speech with the intent of finding a quantitative characterization which could be applied across speaking conditions. To this end, we analyzed the data using the “topline” modeling procedure, as well as peak, valley, and all‐points rms lines, computed b...

The vowels of American English are specified by particular formant frequency patterns and durations. Acoustic data derived from the speech of children between the ages of 69 and 145 weeks show that the vowels that they produce in meaningful words often have mismatched formant frequency patterns and durations. The formant frequency pattern of an intended [i], for example, may have the duration of an [1]. Psychoacoustic experiments using computer edited stimuli derived from the children's vowel productions demonstrate the interaction of formant frequency patterns and duration as acoustic cues to vowel perception. Intrinsic vowel duration and control of the supralaryngeal vocal tract appear to be independent parameters in the acquisition of speech. Hypothetical phonetic features like Tenseness thus can not be causally linked to relative vowel duration.

A Japanese mother and her infant son were recorded on high quality equipment in the home setting. The imitative utterances of the infant were analyzed acoustically and compared to the mother's. The intonation contours from age five weeks were found to reproduce acoustic characteristics of F0 frequency, pattern, and duration; apparently in that order of acquisition. It was also found that the mother raised her F0 from an average of 185 Hz to an average of 325 Hz when interacting with her child. 325 Hz is within the infant's own F0 range, apparently facilitating imitation. At approximately 11 weeks the child also began to spontaneously produce vowel sounds, and to imitate his mother's vowels. The infant vowels were frequency scaled, and not simple acoustic matches. The child could have reproduced some of his mother's vowels almost exactly, but instead finds a phonetic equivalence in an acoustic vowel space that is appropriate to his smaller supralaryngeal vocal tract. [Work supported by NICHHD.]

Recent theories for sentence intonation have proposed quantitative metrics for the F0 contours of sentences. Declination theories have claimed that (1) the “top line” of F0 peaks falls throughout a sentence, (2) the top line peaks at some specified time and then falls. A test of these theories is described using as a corpus the F0 contours that were published in the Lebendige Sprache monograph series between 1937 and 1940. These experimental studies derived F0 contours for speakers of British (RP) and American English reading stories, delivering political speeches, in theatrical performances, and in conversations. Neither claim (1) nor (2) is characteristic of most sentences in the corpus. Although F0 declinations occurred for some sentences for some speakers, other sentences were best characterized by relatively level F0 contours marked by F0 peaks corresponding to phonetic prominence and abrupt, sentence‐terminal F0 falls. In contrast to the claims of declination theories, F0 contours that gradually rose throughout a declarative sentence occurred frequently. Certain F0 phenomenona appear to be “discourse” effects that introduce “new topics.”

In the late babbling and early word stages of development, the vowels that children produce are not as “well formed” as those of adults. Taking into consideration the higher formant frequencies of children's vowels, which follow from their shorter supralaryngeal vocal tracts, one finds, for example, that F1 of an [i] is higher than it should be relative to F2 at ages of approximately 70 weeks. This is presumably a consequence of the shorter length of the child's pharynx relative to the oral cavity (King, 1952). As the child matures, the relative position of F1 and F2 of [i] come closer to the adult model. In addition, there is a greater variability in the formant frequencies of all vowels at earlier ages. This may be a consequence of the child's pharynx not being equal in length to the oral cavity, making vowels like [i] less characteristically “quantal” (Stevens, 1952). This variation may also be due to a lack of motor control. In this study, the relative importance of quantalness versus motor control is...

The present study investigates the relation between speech coarticulation and syntactic processing in children. Specifically, acoustic and perceptual measurements of anticipatory labial and lingual coarticulation are compared to the syntactic abilities of children, as measured by the Rhode Island Test of Language Structure. Two children with good (greater than 95% correct) and two children with poor (less than 85% correct) syntactic comprehension scores produced five tokens each of the following CV syllables ([si], [su], [zi], [zu], [ti], [tu], [di], [du], and [ki], and [ka]). The effects of the vowel on the preceding consonant were acoustically assessed by measuring both second formant and characteristic spectral prominence regions in the consonantal stimuli. Perceptual measurements were then performed by presenting these excised consonantal segments for vowel identification. The correlation between the coarticulation data and the measurements of the children's syntactic abilities is evaluated. The resul...

This experiment examines vowel context effects. In adults, the general spectral characteristics of velar stop consonants vary as a function of vowel context. A velar stop preceding a front vowel has a predominant spectral peak in the mid‐frequency region whereas a velar stop preceding a back or central vowel has a spectral peak in the low‐frequency region and a secondary spectral peak in the high‐frequency region. In the present experiment, CV syllables ([ki], [ka]) were produced by adults and children. Short time spectra were computed for each stimulus. The results indicate that the adult stimuli, as expected, show significant vowel context effects. In the child data, however, the differences for the velar stops preceding front and back vowels are not always present. A perceptual analysis was then conducted using the aperiodic portion of the [ki] and [ka] syllables. These data show that the vowel context effects for adults are perceptually salient but that these cues are often not perceptible in the child stimuli. The results arc discussed in terms of their implications for language learning.

We propose a theory that appears to characterize quantitatively those aspects of the temporal pattern of fundamental frequency (F0) variation that delimits the constituent structure of speech, most notably sentences. The anchor line frequency theory we propose provides a major refinement of the breath‐group theory [P. Lieberman, Intonation, Perception and Language (MIT Press, Cambridge, MA, 1967); J. E. Atkinson, J. Acoust. Soc. Am. 63, 211–222 (1978)]. We will present an analysis of data that is consistent with the hypothesis that the F0 contour (1) provides continuity across voiced‐unvoiced segmental boundaries, (2) that local F0 minima tend to fall on anchor lines, (3) that the anchor lines are linear level, falling, or rising, and (4) that the level anchor is most characteristic for normal, nonemphatic declarative sentences of American English and English. We further propose that the F0 continuity inherent in the anchor lines may play a crucial function in the perception of speech by “tagging” a speak...

Pitch pulses were electronically derived from the utterances of six male speakers of American English who each read eight neutral test sentences in certain “emotional” modes, i.e., as a question, an objective statement, a fearful utterance, a happy utterance, etc. A fixed vowel POVO type synthesizer was excited by these pitch pulses and the resulting pitch contours recorded. The sets of test sentences and pitch contours were each placed in random sequences and separately categorized by linguistically naive listeners in forced judgment tests. Certain of the modes were not easily identified when the sole acoustic cue was the pitch contour. The results of the psychoacoustic test were then correlated with sonagraphic information.

Intonation can be analyzed, in part, in terms of the phonologic feature breath group. The results of an experiment are discussed in which subglottal air pressure and lung air volume were measured as functions of time for four speakers, in addition to the acoustic signal. These data show that the unmarked or “normal” breath group of American English is characterized at the articulatory level by a relatively steady tension of the laryngeal muscles throughout a single‐bounded expiration of air from the lungs. At the end of phonation, the subglottal air pressure falls and hence the fundamental frequency also decreases repidly. The intonation pattern that is a consequence of this breath group is, of course, the one that has been described as characteristic of declarative sentences and numerous other languages. The marked breath group is differentiated at the articulatory level by an increase in the tension of the laryngeal muscles at the end of phonation, which counteracts the falling subglottal air pressure t...

The explanatory powers of the motor theory of speech perception (in the form elaborated at Haskins Laboratories) are enhanced when acoustic factors like those discussed by Stevens in his quantal theory are incorporated into a unified theory. The categorical discrimination of stop consonants may follow from neural feature detectors that have evolved to match the quantal effects of the human supralaryngeal vocal tract. Recent experimental data indicate that the primate auditory system is plastic, i.e., neural “feature detectors” can be set up for auditory events that are important to the animal. The decoding of speech in man appears to be the result of arrays of feature detectors matched to the quantal constraints of the supralaryngeal vocal tract. The effects of hemispheric lateralization indicate that active modeling of the vocal tract by an “analysis‐by‐synthesis” model does not appear to be the correct neural implementation. The unified theory is consistent with a phylogenetic model for the evolution of human speech production and perception. The evolution of the human supralaryngeal vocal tract probably preceded the evolution of the neural mechanisms necessary for decoding human speech.

The object of this experiment was to determine the discriminability of missing pitch pulses in synthesized vowel‐like sounds. Pitch pulses were derived from samples of real speech. Individual pitch pulses were removed from certain parts of these pulse trains. A fixed vowel POVO‐type synthesizer was then excited by these pulse trains, and listeners were asked, in an A–B test, to discriminate between the POVO excited by complete pulse trains and the POVO excited by pulse trains having missing pitch pulses. It was found that a single missing pulse in the middle of a sustained monotone could be easily discriminated. However, stimuli having missing pitch pulses near the onset of the utterance could not be discriminated from stimuli generated from complete pulse trains. Discriminability depended on the position of the missing pulse in the pulse train. Discriminability of missing pitch pulses also depended on the average pitch period, the rate of change of pitch periodicity, and the total stimulus duration.

A system developed by Trager and Smith for the transcription of pitch and stress is in wide use by linguists. Four pitch levels, four stress levels, and three “terminal contours” are used to differentiate special prosodic “morphemes,” which are supposed to carry specific meanings that are independent of the sentences' words. This experiment shows that these “phonemic” elements may have no acoustic correlates; the linguist infers their presence from the words of the sentence, the schema of the Trager‐Smith system, and his intuitive knowledge of the language's structure. The words convey the sentence's meaning to the linguist who then “fills in” the pitch and stress levels appropriate to the prosodic “morphemes” that have that meaning. A set of sentences was transcribed by two skilled linguists. The sentences were electronically processed so that the words were removed, leaving only the contours and stresses. (A POVO was excited with the pitch and amplitude information extracted from the original sentences.) The transcriptions of the POVO output differed markedly from the transcriptions of the complete sentences, except for the primary stresses and final pitch contours. Control procedures confirmed that the linguists were transcribing the actual acoustic signal for the POVO output.

The vowel repertoire of a rhesus monkey (Macaca mulatta) was explored by means of a computer program that calculated formant frequencies from the area function of the animal's supralaryngeal vocal tract, which was systematically varied within the limits imposed by anatomical constraints. The resulting vowels were compared with human vowels and the vocalizations of nonhuman primates previously analyzed by Lieberman [“Primate Vocalizations and Human Linguistic Ability,” J. Acoust. Sac. Amer. 44, 1574–1584 (1968)]. The results are in accord with the previous study. The computer model indicates that the acoustic “vowel space” of a rhesus monkey is quite restricted as compared to the human range. This limitation is due to an anatomical feature of nonhuman primates, the lack of a pharyngeal region that can change its cross‐sectional area. These animals thus lack the output mechanism that is necessary for the production of human speech. Man's speech‐output mechanism is apparently species specific.

An acoustic analysis of perturbations in pitch, that is, sudden irregularities in the glottal excitation rate, during connected speech, is described. The effects of different vocal tract configurations, over‐all sentence intonation, and different speakers on these discontinuities are noted. The degree of accuracy with which these perturbations must be located in time for the retention of natural quality and talker identification is discussed.

Highly redundant messages were recorded at normal and fast rates of talking. Certain words were gated out of these messages and listening tests were performed on these words. Nonredundant messages were then recorded using these same words in similar phonetic environments. The same words were again gated out and listening tests were again performed. Quite apart from these listening teats, two operational measures of redundancy were derived. One measure was based on a finite‐state model of perception and readers were asked to guess what word n would be after they had read all the words through word n − 1 in a sentence. The second measure was based on the hypothesis that people reserve their final decision on the recognition of each word until they perceive an entire sentence. Readers read an entire sentence in the midst of which a word was represented by a dash. They were then asked to guess what the missing word was. Our index of redundancy was the percent of correct guesses. The results of our listening tests show that the intelligibility of the gated words is inversely proportional to the redundancy index obtained by the second measure of redundancy.

High‐speed motion pictures of the human vocal cords during phonation were taken at a maximum rate of 8800 frames per sec by the techniques of indirect laryngoscopy. The Fastax camera was blimped to isolate its operating noise, and the acoustic waveform was picked up at the lips by an Altec 21B condenser microphone. The acoustic waveform was recorded directly on the film in accurate time relationship to the pictures of the vocal cords by an optical variable‐area track. The area of the glottal opening and the displacement of the vocal cords from each other were measured frame by frame. The relationship between these measurements of the vocal cords' vibratory cycle and the resulting acoustic pressure wave was determined for the onset, middle, and end of phonation. When the time delay between the recorded sound at the lips and the glottis was considered, it was found that maximum sound pressure occurred during the closing phase of the vocal‐cord vibratory cycle. Other aspects of the vibratory cycle—in particu...

Subglottal air pressure was measured during the production of unemotional declarative sentences, yes‐no questions, sentences with emphasized words, and simulated emotional discourse. Pressure was directly measured by means of a tracheal puncture and simultaneously measured indirectly by means of an esophageal balloon. The subglottal air pressure was calculated from the esophageal pressure by compensating for the elastic recoil of the lungs. The calculated subglottal air pressure was found to be within 1% of the directly measured pressure. It was necessary to compensate for the elastic recoil of the lungs when calculating the subglottal pressure from the esophageal balloon even when short sentences were produced. Otherwise, the calculated subglottal air pressure at the end of the utterance was too high. The positive error can be as great as 30% for an utterance having a duration of 2 sec.