Search

This study examined acoustic variation of vowels within speakers across speech tasks. The overarching goal of the study was to understand within-speaker variation as one index of the range of normal speech motor behavior for American English vowels. Ten male speakers of American English performed four speech tasks including citation form sentence reading with a clear-speech style (clear-speech), citation form sentence reading (citation), passage reading (reading), and conversational speech (conversation). Eight monophthong vowels in a variety of consonant contexts were studied. Clear-speech was operationally defined as the reference point for describing variation. Acoustic measures associated with the conventions of vowel targets were obtained and examined. These included temporal midpoint formant frequencies for the first three formants (F1, F2, and F3) and the derived Euclidean distances in the F1-F2 and F2-F3 planes. Results indicated that reduction toward the center of the F1-F2 and F2-F3 planes increased in magnitude across the tasks in the order of clear-speech, citation, reading, and conversation. The cross-task variation was comparable for all speakers despite fine-grained individual differences. The characteristics of systematic within-speaker acoustic variation across tasks have potential implications for the understanding of the mechanisms of speech motor control and motor speech disorders.

The study investigated the perception of Mandarin Chinese vowels of 45 congenital hearing-impaired and 45 normal-hearing children aged 4 to 6. There were 30 children at each age group, half of which were hearing-impaired children. All of the hearing-impaired children received hearing aids or cochlear implants before the age of 5. In a picture identification task, listeners identified the target consonant-vowel word among 2–4 contrastive words. The target words and the other contrastive words differed only in consonants. Each target word represented a concrete object and was spoken by a young female native-Chinese talker. Sixteen of the target words ended with monophthong, twenty-two with diphthong and nine with triphthong. Age showed a significant effect on vowel perception for both groups. Normal-hearing children showed significantly better identification of all three types of vowels than hearing-impaired children at the age of 6, whereas the two groups had comparable performance at age of 4 and 5. For hearing-impaired children, a rapid development of diphthong perception occurred between 4 and 5 years old, while a rapid development of monophthong perception between 5 and 6. The effect of hearing loss severity and the use of hearing aids or cochlear implants will be discussed.

The dialect spoken in the small southern Appalachian community in Western North Carolina, formally a part of the Inland South, represents the most archaic features of Southern English, including monophthongization of the diphthong /ai/ to [a:]. While Southerners produce the monophthong before voiced consonants (“prize”) all across the South, pre-voiceless monophthongization (“price”) is more restricted and still occurs in the Inland South. However, as the local culture is gradually shifting toward new mainstream sociocultural norms, so does the local dialect. To document the sound change in progress, this acoustic study examined the /ai/-monophthongization across several generations of local speakers ranging from 8 to 91 year olds, 58 males and 60 females, using a style shifting paradigm. Older generations produced the monophthong irrespective of speaking style and consonant context. Young adults introduced the diphthongal variant in pre-voiceless context. The actual departure from the monophthong occurre...

This study investigated the acoustic features of vowel production in Mandarin-speaking patients with post-stroke spastic dysarthria. The subjects included 31 native Mandarin-speaking patients with post-stroke spastic dysarthria (age: 33–73 years old) and 40 normal adults in a similar age range (age: 22–68 years old). Each subject were recorded producing a list of 28 Mandarin monosyllables that composed of six monophthong vowels (i.e., /a, o, ɣ, i, u, y/) embedded in the /CV/ context. The patients’ speech samples were evaluated by two native Mandarin speakers. The evaluation scores were then used to classify each patient into one of the two categories: mild or moderate-to-severe severity. Midpoint F1 and F2 of each vowel token were extracted and normalized. Results showed no significant differences between the patients and normal speakers on vowel duration. However, the vowel categories in the patients were more scattered and greatly overlapped than in the normal speakers. The magnitude of the vowel disper...

The extent to which it is necessary to model the dynamic behavior of vowel formants to enable vowel separation has been the subject of debate in recent years. To investigate this issue, a study has been made on the vowels of 132 Australian English speakers (male and female). The degree of vowel separation from the formant values at the target was contrasted to that from modeling the formant contour with discrete cosine transform coefficients. The findings are that, although it is necessary to model the formant contour to separate out the diphthongs, the formant values at the target, plus vowel duration are sufficient to separate out the monophthongs. However, further analysis revealed that there are formant contour differences which benefit the within-class separation of the tense/lax monophthong pairs.

The vowel /■/ in Mandarin has traditionally been documented as a simple vowel although it has been noticed to involve more complicated articulator movements in its production than a monophthong. Three male speakers of Beijing Mandarin were recorded for an acoustic study of this vowel. The frequency and energy values of F1 through F4 of the beginning part (around 20 ms), the first quarter, and the rest of the vowel were measured from the spectrogram and LPC using CSL. Results of two‐tail paired t tests show that the first quarter of the vowel has significantly weaker energy in F2 through F4 than the rest of the vowel at p=0.05 level, indicating approximantlike properties in this part. Also, the initial 20 ms of the vowel has higher F1 and lower F2 than the following part, indicating a possible pharyngeal constriction. So phonetically, /■/ consists of a very short pharyngeal approximant [■], a velar approximant [■‖], and a real vowel [■] in sequence. This phonetic characteristic accounts for the fact that glottal stop insertion before a syllable‐initial /■/ is much less likely than that before a real onsetless monophthong /a/, but patterns with that before a syllable‐initial /wo/.

The effect of diminished auditory feedback on monophthong and diphthong production was examined in postlingually deafened Australian-English speaking adults. The participants were 4 female and 3 male speakers with severe to profound hearing loss, who were compared to 11 age- and accent-matched normally hearing speakers. The test materials were 5 repetitions of hVd words containing 18 vowels. Acoustic measures that were studied included F1, F2, discrete cosine transform coefficients (DCTs), and vowel duration information. The durational analyses revealed increased total vowel durations with a maintenance of the tense/lax vowel distinctions in the deafened speakers. The deafened speakers preserved a differentiated vowel space, although there were some gender-specific differences seen. For example, there was a retraction of F2 in the front vowels for the female speakers that did not occur in the males. However, all deafened speakers showed a close correspondence between the monophthong and diphthong formant movements that did occur. Gaussian classification highlighted vowel confusions resulting from changes in the deafened vowel space. The results support the view that postlingually deafened speakers maintain reasonably good speech intelligibility, in part by employing production strategies designed to bolster auditory feedback.

Formant measurements are commonly extracted at fixed fractions across a vowel's duration (e.g., the 1/2 point for a monophthong and the 1/3 and 2/3 points for a diphthong). This approach tacitly relies on the convenience assumption that a speaker always maximally approximates the intended acoustic target at roughly the same point across a vowel's duration. The present paper proposes an alternate method whereby every formant point sampled within a vowel is considered as a possible "pivot" (i.e., turning point), with monophthongs modeled as having one pivot and diphthongs modeled as having two pivots. The optimal pivot for the vowel is then determined by fitting regression lines to the formant trajectory and comparing the goodness-of-fit of these lines to the raw formant data. When applied to a corpus of an American English dialect, the resulting measurements were found to be significantly correlated with previous methods. This suggests that the aforementioned convenience assumption is unnecessary and that the proposed model, which is more faithful to our understanding of articulatory dynamics, is a viable alternative. Moreover, rather than being assumed a priori, the location of the measurement can be treated as an empirical question in its own right.

Previous research suggested a large inter-subject variability in dynamic pitch perception among older individuals (Souza et al., 2011). Although data from younger listeners with normal hearing indicate temporal and spectral variations in complex formant characteristics may influence dynamic pitch perception (Green et al., 2002), the present study examines this interaction in an aging population. The stimulus set includes two monophthongs that have static formant patterns and two diphthongs that have dynamic formant patterns. The fundamental frequency at the midpoint in time of each vowel is kept consistent, while the ratio of start-to-end frequency varies in equal logarithmic steps. Older adults with near-normal hearing are tested using an identification task, in which they are required to identify the pitch glide as either “rise” or “fall.” An experimental task of AX discrimination is also included to verify the identification data. Results to date show inter-subject variability in dynamic pitch perception among listeners with good static pitch perception. Better pitch glide perception with monophthong than diphthong is observed in those individuals who perform poorly in general. The findings suggest a connection between individual abilities to perceive dynamic pitch and to extract the cues from fundamental and formant frequencies. [Work supported by NIH.]

This study tests the hypothesis that the acoustic difference between [a] in English diphthongs (e.g. [a] in "pie'd") and its corresponding monophthong (e.g. [a] in "pod") results from the same pharyngeal gesture being truncated by the following palatal glide in the diphthongal environment. Production data were collected with real-time MRI and have been analyzed using the direct image analysis (DIA) technique, which infers tissue movement by tracking pixel intensity change over time in regions of interest. Preliminary results show that (1) DIA is capable of capturing the timing and magnitude of the pharyngeal constriction gesture which produces [a]; and (2) the proposed hypothesis is supported: the formation time of pharyngeal constriction strongly correlates with the resulting constriction degree as predicted.

Vowel‐inherent spectral change (VISC) refers to vowel‐intrinsic formant movement across a vowel steady state. VISC has been shown to (1) be consistent across talkers within a given dialect, (2) vary regularly across vowels within a dialect, (3) vary regularly across dialects, and (4) be necessary for peak vowel‐identification accuracy. Hence, VISC has become accepted as a phonetic feature of monophthong vowels of North American English. VISC is typically portrayed using averages across tokens and talkers, highlighting regularity but potentially masking individual differences. To understand vowel production by second‐language learners, we were particularly interested in such individual variation. In analyzing individual differences for neighboring target vowels, we found no single time point at which all sets of target vowel tokens were well distinguished from one another. However, looking across three time points, all native‐speaker vowel sets were well distinguished from each possible neighbor set at some time point. Thus, VISC can be seen as the steps in a sort of dance, as each vowel moves to avoid overlapping with another, ultimately causing overlap with another and then more movement. This perspective is compatible with models of efficient coding and stochastic and/or exemplar based models of speech production and perception. [NIH‐NIDCD #1R03DC005561‐01A1.]

An objective analysis has been performed on all 15 Dutch vowels pronounced in /hVt/ words by nine native Dutch, nine non‐native, and six deaf males. Spectral representations of the vowel segments were created by determining the mean output levels of a bank of 16 filters (90–7200 Hz), with 1/3‐oct bandwidths and logarithmic spacing of their center frequencies. The adequacy of the objective analysis is determined by the extent to which spectral information provides an accurate description of pronunciation quality. Spectral distances between the 24 utterances of each monophthong agree rather well with subjective distances obtained by listeners in an elaborate paired‐comparisons experiment. For the various monophthongs, the correlation coefficients are within the range 0.63 to 0.88; averaging across all 12 monophthongs of each speaker results in a coefficient of 0.94. Furthermore, it appeared that the objective spectral analysis is as reliable as a subjective assessment by magnitude estimation by two to three listeners. Using principal components analysis (PCA), the number of dimensions by which the vowel spectra are described can be reduced. For the various monophthongs the range of the correlation coefficients between subjective distances and objective distances in a two‐dimensional PCA subspace is 0.30–0.93. The three groups of speakers can still be distinguished in this subspace. In the extreme case of the deaf speakers all vowels are strongly ‘‘neutralized,’’ whereas the different vowels of the native speakers are well separated, especially after speaker normalization; results are less clear for the non‐natives.

This study focuses on acoustic characteristics of vowels in the regional variety of Southern American English spoken in western North Carolina. It examines the cross‐generational pattern of positional vowel changes with special reference to vowel‐inherent spectral change. 140 speakers, males and females, ranging in age from 8‐year‐old children to adults in their early 90s produced 13 vowel categories in isolated hVd words. All participants were born and raised in the area and speak the local dialect. The results show that the reorganization of the vowel system occurs gradually, generation by generation, and there is a correspondence between positional vowel changes and the changes in formant dynamics. In each age group, females represent a more advanced stage of vowel change compared to males. These data clearly document both the systemic nature of vowel change and close interdependent relationships between the nominal monophthongs and diphthongs. The categorical distinction between monophthong and diphthong changes across generations. The oldest speakers produce monophthongal diphthongs and diphthongal monophthongs, whereas younger speakers show increased monophtongization of the former and diphthongization of the latter. [Work supported by NIH.]

In order to test the correspondence between variability in fricative production and variability in the fricative spectrum, the spectral center of gravity (COG) is compared with the EPG center of gravity measured along the horizontal dimension. The CV coarticulatory effects in particular are examined. Data are taken from CV tokens produced by four female speakers of Australian English. Tokens consisted of the coronal fricatives /T s S D z Z/ in seven monophthong vowel contexts. Rounded vowels are excluded so as to minimize spectral changes due to lip‐rounding. The consistency between the articulatory and the acoustic data is particularly striking for the palato‐alveolar fricatives, although results are also highly consistent for the other two fricative places of articulation examined. However, it is not always clear that the variability in fricative production is conditioned by the vowel context. Overall, the sibilant fricatives seem particularly resistant to coarticulation and to variability in production, while the nonsibilant dental shows a good deal of variability, including some conditioning by the vowel context. These results have implications for the hyper‐ and hypo‐theory of speech production. It is suggested that sibilant fricatives do not lend themselves to the articulatory imprecision, which characterizes other perceptually salient, and typologically common, speech sounds.

This study provides new native speaker data regarding the vowel space and duration contrasts of Vietnamese monophthongs. These data add to existing empirical accounts and raise questions about descriptions aimed at adult language learners. In addition, specific acoustic measures target difficulties that English‐speaking adults encounter when producing Vietnamese vowels. These measures include the rounding distinction for back vowels, height distinction for central vowels, and duration distinction between short and long vowels. Results show that adult learners exhibit greater variability and overall less accurate production than native speakers for the high‐back unrounded vowel u', with slightly better performance for the mid‐back unrounded vowel o', perhaps because it is similar to English / Λ /. Additionally, these learners demonstrate varying levels of success with regard to the production of central vowel height but generally fail to distinguish the two vowel duration categories exhibited by native speakers.

Nearey [ICSLP (1992)] suggests using generalized linear modeling (GLM) to investigate the role of vowel‐intrinsic (e.g., role of F0) and vowel‐extrinsic (e.g., z‐transformation, log‐mean transformation) normalization schemes. A generalized linear model consists of two parts: a linear predictor and a nonlinear link function. GLM is used to predict values both for dependent variables with discrete distributions and for variables that are nonlinearly related to the predictors. GLM can therefore be useful for dealing with relations between vowel‐intrinsic and vowel‐extrinsic information and vowel categories, since intrinsic factors can be tested as predictors in the linear predictor function and extrinsic factors can be tested using the link function. The GLM approach was applied to speech data from 80 speakers from four different regions in The Netherlands. The regional accent spoken in each group was different from the other three groups. The nine monophthong vowels of Dutch were produced in a neutral conte...

An attempt was made to evaluate the performance of several methods for speaker normalization in both the acoustic and the perceptual domain of speech. This was done by comparing the acoustic distributions applied to the same vowel data and further by comparing the acoustic distributions to the perceptual distributions of the vowel data. The normalization methods included, among others, extrinsic (e.g., z‐score transformation) and intrinsic methods (Bark transformation), and formant weighting (F2) and correction (F3−F2). To obtain the acoustic distributions, the normalization methods were applied to F0 and formant data from monophthong vowels in /sVs/ context of male and female speakers of Standard Dutch. The perceptual distributions were obtained through an experiment with phonetically trained listeners, whose task was to judge each vowel’s height, place of constriction and amount of rounding/spreading. The acoustic and perceptual distributions were compared using correlational‐ and cluster‐analysis techniques. When describing the results of these tests, the focus will be on the extent to which the variation and overlap are the same in the acoustic and perceptual domains, and which normalization methods show a pattern of overlap and variation most similar in both domains. [Work supported by The Netherlands Organization for Research (NWO).]

A direct classification of nine Thai monophthongs was conducted on two‐dimensional acoustic–phonetic feature spaces. Formant frequencies of a monophthong have been extracted from a stable vowel portion of a syllable nucleus. The F1 and bark‐difference F1‐F0 represents tongue height. The F2 and bark‐difference F3‐F2 represents tongue advancement. The Bayesian classifier utilizes statistical parameters, mean and standard deviation, of each monophthongs in classification. Two‐dimensional feature vectors comprise two acoustic–phonetic features, F2 and F1 in linear, mel, bark, and also, bark‐difference F3‐F2 and F1‐F0. The classification results on the F2 and F1 space are 86.3325% in linear, 84.6187% in mel, 84.2331% in bark, 79.3916% in bark‐difference F2‐F1 and F1‐F0, and also 79.9914% in bark‐difference F3‐F2 and F1‐F0. Considering confusion matrices, the high vowels (/i:, v:, u:/), front‐middle vowel (/e:/), and center‐middle vowel (/q:/) have higher results than other groups which resulted from smaller ov...

A computer‐based system which provides real‐time feedback for vowel articulation training for the hearing impaired is described. This system is a revised version of the training aid described in previous papers [Zahorian and Correal, J. Acoust. Soc. Am. 95, 3014(A) (1994); Beck and Zahorian, ICASSP‐92, II–241‐244]. Revised feature extraction and classification algorithms improve accuracy and allow processing of vowels spoken in CVC contexts. Cost has been reduced by a new Windows 95/NT‐based implementation which requires no specialized DSP or audio hardware. Steady‐state vowel and CVC tokens have been collected from 24 adult male, 28 adult female, and 46 child speakers to provide training data for semi‐speaker‐independent neural network classifiers used in the system. Displays include a two‐dimensional F1/F2 style display which provides continuous feedback, and a vowel bar graph display. Each accommodates four speaker groups: adult female, adult male, child, and general. Tests with ten steady‐state monophthong vowels produced by speakers outside the training set indicate that typically over 80% of vowels are correctly depicted by the bar graph display and over 75% by the 2‐D display. A third display for vowels spoken in CVC contexts is under development. [Work funded by NSF, Grant No. NSF‐BES‐9411607.]

The purpose of this study was to replicate and extend the classic study of vowel acoustics by Peterson and Barney (PB) [J. Acoust. Soc. Am. 24, 175-184 (1952)]. Recordings were made of 45 men, 48 women, and 46 children producing the vowels /i,I,e, epsilon,ae,a, [symbol: see text],O,U,u, lambda,3 iota/ in h-V-d syllables. Formant contours for F1-F4 were measured from LPC spectra using a custom interactive editing tool. For comparison with the PB data, formant patterns were sampled at a time that was judged by visual inspection to be maximally steady. Analysis of the formant data shows numerous differences between the present data and those of PB, both in terms of average frequencies of F1 and F2, and the degree of overlap among adjacent vowels. As with the original study, listening tests showed that the signals were nearly always identified as the vowel intended by the talker. Discriminant analysis showed that the vowels were more poorly separated than the PB data based on a static sample of the formant pattern. However, the vowels can be separated with a high degree of accuracy if duration and spectral change information is included.

Recent investigations suggest that analysis of spontaneous speech, rather than citation form speech, may have greater explanatory power for models of speech production and perception. The present study sought to establish a preliminary database of selected acoustic characteristics of vowels produced by women. Acoustic analyses were performed on the spontaneous and citation form speech of five elderly and five young female subjects. Measurements included segment durations and F1 and F2 temporal midpoints for monophthong vowels, and formant trajectory characteristics for selected diphthongs. Vowels were grouped for analysis by syllabic and phrase level stress and by phonemic environment. Acoustic characteristics of vowels will be illustrated by F1–F2 plots and distribution histograms. Results will be discussed relative to speaker characteristics, speaking condition, and how these acoustic measures may relate to speech motor control.

The effects of noise and reverberation on the identification of monophthongs and diphthongs were evaluated for ten subjects with moderate sensorineural hearing losses. Stimuli were 15 English vowels spoken in a /b-t/ context, in a carrier sentence. The original tape was recorded without reverberation, in a quiet condition. This test tape was degraded either by recording in a room with reverberation time of 1.2 s, or by adding a babble of 12 voices at a speech-to-noise ratio of 0 dB. Both types of degradation caused statistically significant reductions of mean identification scores as compared to the quiet condition. Although the mean identification scores for the noise and reverberant conditions were not significantly different, the patterns of errors for these two conditions were different. Errors for monophthongs in reverberation but not in noise seemed to be related to an overestimation of vowel duration, and there was a tendency to weight the formant frequencies differently in the reverberation and quiet conditions. Errors for monophthongs in noise seemed to be related to spectral proximity of formant frequencies for confused pairs. For the diphthongs in both noise and reverberation, there was a tendency to judge a diphthong as the beginning monophthong. This may have been due to temporal smearing in the reverberation condition, and to a higher masked threshold for changing compared to stationary formant frequencies in the noise condition.

To evaluate the relationship between physical characteristics of the lips during vowel production and vowel lipreading confusions, four female talkers were videotaped producing 15 American English vowels and diphthongs in an /h/-V-/g/ context. Ten normal-hearing adults identified the stimuli through lipreading. Three analyses were performed. First, using confusion matrices for individual and pooled talkers, the stimuli were located in a two-dimensional space using multidimensional scaling. The ten monophthongs revealed a clear lip spreading/rounding dimension and a tongue height dimension, and while diphthongs also showed influence of lip rounding, more variability on the tongue height dimension was apparent. Second, tracings were made of the talkers' lips on a single videotape field representing the maximum opening or constriction for each of the 40 monophthong tokens (ten vowels X four talkers), and six physical measurements of the tokens were derived as descriptors of the vowel nuclei. Third, difference scores and other measures of physical pairwise similarity were used as predictors of two ways of representing the vowel lipreading confusions in a multiple regression paradigm. Results indicated that the physical measures were moderately successful as predictors of vowel perception (accounting for approximately 50% of the variance in the perceptual distance measure), although considerable differences in the strength of the prediction occurred among talkers.