X-ray Microbeam Speech Production Database Research Articles

Speech Production in Two Occlusal Classes

Background: The influence of the occlusal class in speech production has been studied using the X-ray Microbeam Speech Production Database (XRMB-SPD). Objectives/aims: This study aimed to relate the occlusal classes I and II with speech production adaptations. Methods: The Modified A-Space method was used to select 4 speakers (1 male and 1 female class I, 1 male and 1 female class II). Articulatory and acoustic features of the vowels were studied using different tasks and methods. The articulatory and acoustic features of consonants in male and female speakers of class I and class II from the XRMB-SPD were also described in detail. Measures extracted from multitaper spectra and articulatory data were used, to observe individual differences related with gender and dental occlusion. Results: Results showed some structural differences related to occlusal class and variance in class II subjects’ structures and articulatory adaptations. However, subjects showed a high adaptation capacity, being able to adjust their articulations to produce all vowels. Conclusions: Speech production variability is related with orofacial structures’ variance. Different structures produce various functional adaptations and distinct speech signals.

日英母語話者による英語弱化母音の音響・調音特徴 : X線マイクロビームデータベースに基づく分析

日英母語話者による英語弱化母音の音響・調音特徴 : X線マイクロビームデータベースに基づく分析

Recovery of articulatory information from acoustics using linear mixed effects models

The current study attempts to establish a lower bound on the degree to which articulatory information can be recovered from the speech signal. The recovery of articulatory information is modeled with comparatively simple linear mixed-effects models using synchronized articulatory and acoustic data from the University of Wisconsin X-Ray Microbeam Speech Production Database (Westbury, 1994). The performance of this model is compared with other speech inversion models. While previous work [Mitra etal. (2010) and others] demonstrates that a more sophisticated model can perform speech inversion quite accurately, the results here show that even a strongly constrained linear model performs better than might be expected given that the well-known non-linearity of the acoustic-articulatory mapping. More detailed investigation of the results indicates that while non-linearity in the acoustic-articulatory mapping affects the recovery of articulatory information for all tongue regions, it contributes considerably more to error in recovering tongue-tip position. For the tongue body, the correlation between predicted and observed tongue body position is approximately 0.9 while the correlation for tongue tip position was approximately 0.65. A linear approximation seems to be sufficient to recover tongue body position with average error of approximately 12%, even with limited acoustic information.

Vowel constrictions are recoverable from formants

The area function of the vocal tract in all of its spatial detail is not directly computable from the speech signal. But is partial, yet phonetically distinctive, information about articulation recoverable from the acoustic signal that arrives at the listener's ear? The answer to this question is important for phonetics, because various theories of speech perception predict different answers. Some theories assume that recovery of articulatory information must be possible, while others assume that it is impossible. However, neither type of theory provides firm evidence showing that distinctive articulatory information is or is not extractable from the acoustic signal. The present study focuses on vowel gestures and examines whether linguistically significant information, such as the constriction location, constriction degree, and rounding, is contained in the speech signal, and whether such information is recoverable from formant parameters. Perturbation theory and linear prediction were combined, in a manner similar to that in Mokhtari (1998) [Mokhtari, P. (1998). An acoustic-phonetic and articulatory study of speech-speaker dichotomy. Doctoral dissertation, University of New South Wales], to assess the accuracy of recovery of information about vowel constrictions. Distinctive constriction information estimated from the speech signal for ten American English vowels were compared to the constriction information derived from simultaneously collected X-ray microbeam articulatory data for 39 speakers [ Westbury (1994). Xray microbeam speech production database user's handbook. University of Wisconsin, Madison, WI]. The recovery of distinctive articulatory information relies on a novel technique that uses formant frequencies and amplitudes, and does not depend on a principal components analysis of the articulatory data, as do most other inversion techniques. These results provide evidence that distinctive articulatory information for vowels can be recovered from the acoustic signal.

Acoustic and Articulatory Features of Diphthong Production: A Speech Clarity Study

The purpose of this study was to evaluate how speaking clearly influences selected acoustic and orofacial kinematic measures associated with diphthong production. Forty-nine speakers, drawn from the University of Wisconsin X-Ray Microbeam Speech Production Database (J. R. Westbury, 1994), served as participants. Samples of clear and conversational productions of the word combine were extracted for analysis. Analyses included listener ratings of speech clarity and a number of acoustic and articulatory kinematic measures associated with production of the diphthong /aI/. Key results indicate that speaking clearly is associated with (a) increased duration of diphthong-related acoustic and kinematic events, (b) larger F1 and F2 excursions and associated tongue and mandible movements, and (c) minimal evidence of change in formant transition rate. Overall, the results suggest that clarity-related changes in diphthong production are accomplished through larger, longer, but not necessarily faster diphthong-related transitions. The clarity-related adjustments in diphthong production observed in this study conform to a simple model that assumes speech clarity arises out of reduced overlap of articulatory gestures.

Acoustic-articulatory mapping in vowels by locally weighted regression

A method for mapping between simultaneously measured articulatory and acoustic data is proposed. The method uses principal components analysis on the articulatory and acoustic variables, and mapping between the domains by locally weighted linear regression, or loess [Cleveland, W. S. (1979). J. Am. Stat. Assoc. 74, 829-836]. The latter method permits local variation in the slopes of the linear regression, assuming that the function being approximated is smooth. The methodology is applied to vowels of four speakers in the Wisconsin X-ray Microbeam Speech Production Database, with formant analysis. Results are examined in terms of (1) examples of forward (articulation-to-acoustics) mappings and inverse mappings, (2) distributions of local slopes and constants, (3) examples of correlations among slopes and constants, (4) root-mean-square error, and (5) sensitivity of formant frequencies to articulatory change. It is shown that the results are qualitatively correct and that loess performs better than global regression. The forward mappings show different root-mean-square error properties than the inverse mappings indicating that this method is better suited for the forward mappings than the inverse mappings, at least for the data chosen for the current study. Some preliminary results on sensitivity of the first two formant frequencies to the two most important articulatory principal components are presented.

Recovering area functions from acoustics and midsagittal profiles.

Simultaneously recorded midsagittal pellet positions and acoustic output for several subjects comprise the Wisconsin X‐Ray Microbeam Speech Production Database (XRMB‐SPD). An articulatory synthesizer, EASY, was used to help recover area function parameters corresponding to the production of vowels by subjects in the XRMB‐SPD by means of analysis‐by‐synthesis. Specifically, for each subject, an initial personalized midsagittal profile was created based on the pellet positions, pellet trace, and pharyngeal wall trace during neutral vowel production. Based on the scales of this initial profile for a subject, subsequent midsagittal profiles were generated from the pellet data during various vowel productions. Each midsagittal profile and the corresponding first three formant frequencies were used to optimize the parameters of an area function in the EASY for closest acoustic match. The optimization was performed with a genetic algorithm. The results are examined in terms of consistency within a subject for different vowels and consistency within vowel for different subjects. Synthesized examples of vowel production will also be presented. [Work partially supported by Grant DC‐001247 to CReSS LLC.]

Vowel recognition from articulatory position time‐series data.

The goal of this project is to determine the accuracy and processing speed of different approaches for mapping time‐varying articulatory positional data to vowels. Three widely used classifiers were compared on two datasets: one single speaker and one multiple speaker. The single‐speaker dataset was acquired using the Articulograph AG500. The multiple‐speaker dataset was obtained from seven speakers in the Xray Microbeam Speech Production Database (Westbury, 1994). The recognition rate for single speaker dataset (eight English vowels) ranged from 94.25% to 98.1%, and from 62.38% to 99.35% for the multiple‐speaker dataset. For the single‐speaker dataset, recognition accuracy was comparable across classifiers. For the multiple‐speaker dataset, recognition accuracy was better for the Support Vector Machine and C4.5 than for the neural networks. The decision tree generated by C4.5 was consistent with articulatory features commonly used to descriptively distinguish vowels. Moreover, the Support Vector Machine and C4.5 performed much faster than did the neural network. The high recognition rates observed suggest that static vowels can be accurately recognized from articulatory position time‐series data. The findings are intended to improve the accuracy and response time of a real‐time articulatory‐movement based speech synthesizer.

Secant lines in a preliminary study of phonetic reduction of / ɪ / and /ε /

Two subjects from the X-Ray Microbeam Speech Production Database were examined in their production of the vowels /I/ and /epsilon/ in alveolar and dental consonant contexts. Secant lines, or first-order splines, between the three most anterior pellets were examined at vowel critical times. These critical times were zero crossings in the tangential acceleration of the midpoints of the secant lines. We expected and found, in general, that vowel reduction occurred as a function of vowel duration in measures of the secant line midpoint-to-palate distance and secant line orientation at vowel critical times. The shorter the vowel, the smaller the distance of the secant line midpoints to the palate and the less downward the orientation of the secant lines at the vowel critical times. Phonetic reduction was also apparent in the formant frequencies. There were differences between the speakers in terms of the range of vowel duration and degree of reduction. The subjects differed in the functional parts of the tongue spanned by the secant lines and the shape of their palates. These differences were factors in the observed relations between formant frequencies and the articulatory, secant line measures for each subject.

Time-varying acoustic and articulatory characteristics of American English [ɹ]: a cross-speaker study

The present study is an attempt to discover time-varying acoustic correlates of articulatory movements of American English [ɹ] using the X-ray microbeam speech production database. American English [ɹ] is known to be variable in its articulatory postures, whereas a single acoustic correlate—a low third formant—is associated with the sound. A previous study (Westbury, Hashi, & Lindstrom (1998), Speech Communication 26, 203–226) found no relation between articulatory postures and formant frequencies at phonation onset of words beginning with the sound. The present study examined acoustic–articulatory relations in time-varying acoustic and articulatory characteristics in the transition from [ɹ] to a following low- or mid-back vowel. In intervals involving a major F3 rise, a relatively strong correlation ( r⩾0.7) was observed between the average rate of change in F3 (F3 slope) and the speed of the tongue pellet closest to the palate at the onset of the major F3 rise. Results from the study are discussed with respect to the acoustic theory of speech production.

Gender-specific articulatory–acoustic relations in vowel sequences

Differences in male and female vocal tract dimensions are hypothesized to have a number of dynamic consequences—differences in target attainment, articulatory speed, and acoustic vowel space dimension. Evidence for some of these predictions is sought by investigating articulatory and acoustic patterns in interword vowel sequences in the University of Wisconsin X-ray Microbeam Speech Production Database ( UW-XRMBDB). Means of formant and lingual pellet tracks throughout such vocalic stretches exhibit similarities in acoustic and articulatory form for male and female groups, but show significant gender-specific differences in both articulatory and acoustic space traversed, with females making greater acoustic excursions for shorter articulatory distances.

Dynamic consequences of differences in male and female vocal tract dimensions.

Phonetic differences between male and female speakers are generally considered in terms of the static acoustic and perceptual consequences of different articulatory dimensions. This article investigates the dynamic acoustic and articulatory implications of differences in mean male and female vocal tract dimensions. The temporal acoustic consequences of time-varying twin-tube resonators of different dimensions are explored, and the possible implications for human speech production are considered. Empirical support for the theoretical predictions is sought by investigating the kinematic and acoustic patterns in diphthong productions from 26 female and 22 male speakers in the University of Wisconsin X-ray Microbeam Speech Production Database. Aside from expected acoustic differences, the shape of male and female formant tracks plotted in Bark space is found to be very similar. Male and female patterns of tongue movement, however, are found to be very dissimilar. The mean male diphthong, defined by the tracks of four midsagittal pellets, is characterized by greater pellet excursions, higher pellet speed, and consistently larger dorso-palatal strictures than its female counterpart. The empirical findings suggest that gender-specific dynamic behavior could be an important factor in accounting for nonuniform vowel system differences, but at the same time having more wide-ranging implications for transitional phenomena and undershoot.

Kinematic event patterns in speech: special problems.

The view that each utterance is fundamentally a pattern of serially-ordered events underlies a group of well-known speech kinematic studies emphasizing temporal coordination among articulators. Methodological problems that might affect the validity and significance of conclusions from these studies are identified. Results from a new analysis of synchronous acoustic and fleshpoint-kinematic data, recorded from 53 normal young-adult speakers of American English, are then reported. The kinematic data represented speech-related actions of the tongue blade and dorsum, both lips, and the mandible, during the test words special and problem, and were drawn from an existing X-ray microbeam speech production database. Distributions of event patterns across speakers revealed four main results: (1) different patterns for the two test words; (2) a comparable degree of cross-speaker agreement about relative tongue and jaw movement timing, but marked disagreement about lip and jaw movement timing, between test words; (3) highly distinctive movement patterns for some speakers; and, (4) a general conclusion that serial event order, alone, provides very limited understanding of movement patterns produced by individual speakers. By design, these results focus attention on methods of kinematic event pattern analyses, and the general value of such analyses for insights about speech production.

Tongue surface dynamics during speech and swallowing

This investigation characterizes tongue surface dynamics that underlie phonemic variation and that distinguish speech from swallowing. Vertical displacements of pellets affixed to the tongue were extracted from the x-ray microbeam database [Westbury, J. X-ray Microbeam Speech Production Database Users Handbook, Version 1 (1994)], which contains articulatory kinematic data from 57 typical speakers. Participants recited 21 vowel–consonant–vowel (VCV) combinations, read the Grandfather Passage, and swallowed 10 cc of water. Consonantal context was manipulated in the VCV utterances as a means to describe the range of tongue dynamics produced during speech. For each task, tongue dynamics was quantified by performing zero-lag cross correlations on selected pellet pairs. A coupling index was then computed by scaling the derived coefficients by a multiplier that reflected the amplitude of each displacement signal. A wide range of movement coupling among tongue pellets was observed across tasks. Phonemic differentiation in vertical tongue dynamics was observed as coupling varied predictably across marker pairs with place of articulation. Moreover, tongue displacements for speech and swallowing clustered into distinct groups based on their coupling profiles. The strengths and weaknesses of the coupling index for characterizing tongue surface dynamics across multiple speakers are considered.

Vocal tract normalization for midsagittal articulatory recovery with analysis-by-synthesis.

A method is presented that accounts for differences in the acoustics of vowel production caused by human talkers' vocal-tract anatomies and postural settings. Such a method is needed by an analysis-by-synthesis procedure designed to recover midsagittal articulatory movement from speech acoustics because the procedure employs an articulatory model as an internal model. The normalization procedure involves the adjustment of parameters of the articulatory model that are not of interest for the midsagittal movement recovery procedure. These parameters are adjusted so that acoustic signals produced by the human and the articulatory model match as closely as possible over an initial set of pairs of corresponding human and model midsagittal shapes. Further, these initial midsagittal shape correspondence need to be generalized so that all midsagittal shapes of the human can be obtained from midsagittal shapes of the model. Once these procedures are complete, the midsagittal articulatory movement recovery algorithm can be used to derive model articulatory trajectories that, subsequently, can be transformed into human articulatory trajectories. In this paper the proposed normalization procedure is outlined and the results of experiments with data from two talkers contained in the X-ray Microbeam Speech Production Database are presented. It was found to be possible to characterize these vocal tracts during vowel production with the proposed procedure and to generalize the initial midsagittal correspondences over a set of vowels to other vowels. The procedure was also found to aid in midsagittal articulatory movement recovery from speech acoustics in a vowel-to-vowel production for the two subjects.

Influences of pellet markers on speech production behavior: acoustical and perceptual measures.

Peri- and intraoral devices are often used to obtain measurements concerning articulator motions and placements. Surprisingly, there are few formal evaluations of the potential influence of these devices on speech production behavior. In particular, the potential effects of lingual pellets or coils used in x-ray or electromagnetic studies of tongue motion have never been evaluated formally, even though a large x-ray database exists and electromagnetic systems are commercially available. The x-ray microbeam database [Westbury, J. "X-ray Microbeam Speech Production Database User's Handbook, version 1" (1994)] includes several utterances produced with pellets-off and -on, which allowed us to evaluate effects of pellets for the utterance, She had your dark suit in greasy wash water all year, using acoustic and perceptual measures. Overall, there were no acoustic or perceptual measures that showed consistent effects of pellets across speakers, but certain effects were consistent either within a given speaker or in direction across a subgroup of the speakers. The results are discussed in terms of the general goodness of the assumption that point parameterization of lingual motion does not interfere with normal articulatory behaviors. A brief screening procedure is suggested to protect articulatory kinematic experiments from those individuals who may show consistent effects of having devices placed on perioral structures.

What do speakers’ tongues do with a suit in greasy wash water?

The sad fact is that little is known about kinematic variability during natural, connected speech. For example, nearly nothing is known about motions of the tongue for one speaker, or across several speakers, repeating the same phrase. Consequently, the generality of common assumptions concerning those motions cannot be judged. The present study describes the spatio-temporal variability of lingual fleshpoint movements during the consonants /t n g/, across five tokens of the phrase ‘‘...suit in greasy...,’’ extracted from a 13-syllable sentence repeated by 20 normal American English speakers. Data were drawn from the University of Wisconsin x-ray Microbeam Speech Production Database. This phrase was chosen because allowable allophones for its step consonants permit a wide range of articulatory possibilities for phonetically naive speakers producing connected speech. Results of graphic analyses of fleshpoint trajectories and time histories show that lingual gestures associated with these consonants vary substantially within and between speakers. Moreover, these gestures sometimes differ markedly from expectations based on analyses of these consonants in nonsense-syllable or citation-word productions. Such findings have implications for synthesis and recognition of connected speech, as well as for theories of gesture-phoneme correspondence. [Work supported by NIH Grant No. DC00820.]