Intra-speaker Variability Research Articles

Urban rejection of the vernacular: The SVS undone

AbstractIn Raleigh, North Carolina, a Southern U.S. city, five decades of in-migration of technology-sector workers from outside the South has resulted in large-scale contact between the local Southern dialect and non-Southern dialects. This paper investigates the speed and magnitude of the reversal of the Southern Vowel Shift (SVS) with respect to the five front vowels, using Trudgill's (1998) model of dialect contact as a framework. The data consist of conversational interviews with 59 white-collar Raleigh natives representing three generations, the first generation having reached adulthood before large-scale contact. Acoustic analysis shows that all vowels shift away from their Southern variants across apparent time. The leveling of SVS variants begins within the first generation to grow up after large-scale contact began, and contrary to predictions, this generation does not show wide inter- or intraspeaker variability. Previous studies of dialect contact and new dialect formation suggest that leveling of regional dialect features and the establishment of stable linguistic norms occurs more quickly when children have regular contact with one another. Dialect contact in Raleigh has occurred primarily within the middle and upper classes, the members of which are densely connected by virtue of schools and heavy economic segregation in neighborhood residence.

Investigating the acoustic and phonetic correlates of deceptive speech

The following study describes an initial investigation into the acoustic and phonetic correlates of deceptive speech using auditory and acoustic analysis. Due to the lack of extant data suitable for acoustic analysis, a laboratory-based experiment was designed which employed a mock-theft paradigm in conjunction with a “security interview” to elicit truthful and deceptive speech as well as control data from a total of ten male native British English speakers. Using Praat, the control, truthful, and deceptive speech samples were analyzed on a range of speech parameters including f0 mean and variability, intensity, vowel formant frequencies, and speaking/articulation rate. Preliminary analysis suggests that truth-tellers and liars cannot be differentiated based on these speech parameters. Not only was there a lack of significant changes for the majority of parameters investigated but also, if change was present it failed to reveal consistencies within and between speakers. The remarkable amount of inter and intra-speaker variability underlines the fact that deceptive behavior is individualized and very multifaceted. As well as providing a basis for future research programs, the present study should encourage researchers and practitioners to evaluate critically what is (im)possible using auditory and machine based analyses with respect to detecting deception from speech.

How knowledge of speech acoustics can improve the robustness of automatic speech recognition

In this talk, we discuss the ways in which the parallel study of speech production and speech perception can help us develop better automatic speech recognition systems. The ultimate goal of speech recognition (recognition of spontaneous speech from any talker in any language) is still elusive due to a high degree of inter- and intra-speaker variability for production of a given sequence of sounds. While the acoustic information required for recognition may be present in the signal, its distribution, strength, and location are consistent and predictable only as a function of lawful changes in speech movements and/or listener perceptions. Understanding speech acoustics from this perspective is vitally important if we are going to achieve our ultimate goal. We will give several examples of lessons learned from studies of speech production and speech perception and how the knowledge gained can inform the engineering of robust ASR systems.

Intra-speaker variability in palatometric measures of consonant articulation

Electropalatometry is a useful clinical and research tool for measuring linguapalatal contact. The goal of this study was to examine intra-speaker variability in performance. Twenty individuals spoke VCV nonsense words using a schwa in the initial position, the 15 palatal consonants, and three corner vowels, /ɑ/, /i/, /u/. A variability index was created to examine speaker consistency. Different aspects of articulation (i.e. place, manner, voicing, coarticulation) were compared. Significant differences for variability were found for place of articulation in the /i/ vowel context and for manner of articulation in the /ɑ/ vowel context. Also for the /ɑ/ vowel, significant differences were found between the commonly misarticulated /l/, /r/, and /s/. The data reveal differences in production variability across sounds within an individual, as well as from one speaker to another. This knowledge of typical performance may guide the interpretation of data from disordered speakers in future studies. Learning Outcomes: The reader will be able to: (1) Describe how articulatory performance and variability can be assessed with the palatometer; (2) Describe which sounds are typically most or least variable in their production; and (3) Understand the significance of individual speaker differences in normal consonant articulation.

Comparison of magnetic resonance imaging-based vocal tract area functions obtained from the same speaker in 1994 and 2002

A new set of area functions for vowels has been obtained with magnetic resonance imaging from the same speaker as that previously reported in 1996 [Story et al., J. Acoust. Soc. Am. 100, 537-554 (1996)]. The new area functions were derived from image data collected in 2002, whereas the previously reported area functions were based on magnetic resonance images obtained in 1994. When compared, the new area function sets indicated a tendency toward a constricted pharyngeal region and expanded oral cavity relative to the previous set. Based on calculated formant frequencies and sensitivity functions, these morphological differences were shown to have the primary acoustic effect of systematically shifting the second formant (F2) downward in frequency. Multiple instances of target vocal tract shapes from a specific speaker provide additional sampling of the possible area functions that may be produced during speech production. This may be of benefit for understanding intraspeaker variability in vowel production and for further development of speech synthesizers and speech models that utilize area function information.

Unsupervised intra-speaker variability compensation based on Gestalt and model adaptation in speaker verification with telephone speech

In this paper, an unsupervised intra-speaker variability compensation (ISVC) method based on Gestalt is proposed to address the problem of limited enrolling data and noise robustness in text-dependent speaker verification (SV). Experiments with two databases show that: ISVC can lead to reductions in EER as high as 20% or 40% and ISCV provides reductions in the integral below the ROC curve between 30% and 60%. Also, the observed improvements are independent of the number of enrolling utterances. In contrast to model adaptation methods, ISVC is memoryless with respect to previous verification attempts. As shown here, unsupervised model adaptation can lead to substantial improvements in EER but is highly dependent on the sequence of client/impostor verification events. In adverse scenarios, such as massive impostor attacks and verification from alternated telephone line, unsupervised model adaptation might even provide reductions in verification accuracy when compared with the baseline system. In those cases, ISVC can even outperform adaptation schemes. It is worth emphasizing that ISVC and unsupervised model adaptation are compatible and the combination of both methods always improves the performance of model adaptation. The combination of both schemes can lead to improvements in EER as high as 34%. Due to the restrictions of commercially available databases for text-dependent SV research, the results presented here are based on local databases in Spanish. By doing so, the visibility of research in Iberian Languages is highlighted.

Intraspeaker variability in vowel production: An investigation of motherese, hyperspeech, and Lombard speech in Jamaican speakers

Examination of five acoustic parameters (F0, F1, F2, segmental duration, and intensity) allowed comparison of Infant-Directed Speech (IDS), Hyperspeech, and the Lombard reflex. These phenomena have typically been investigated separately, each characterized in terms of one or two acoustic features (e.g., intensity for Lombard speech). The present two-part experiment examined intraspeaker and interspeaker variability in speakers of Jamaican Creole and Jamaican English. IDS and Lombard speech showed similar adjustments in two parameters (F0 and intensity), and IDS and Citation speech showed the greatest spectral differences. This result is important because it shows that while one or two acoustic parameters may be crucial characteristics of a type of exaggerated speech, task production involved the systematic adjustment of a complex set of continuous acoustic parameters. Taken together, the acoustic outcomes for the types of perturbations investigated here allow the tasks to be arranged along a continuum. These adjustments resemble patterns associated with the manipulation of sociolinguistic markers reported in sociophonetic studies of “style-shifting.” We argue that the two phenomena may be treated under a unified account of intraspeaker variability that builds upon the sociolinguistic concept of Audience Design.

Articulation Rate and Vowel Space Characteristics of Young Males With Fragile X Syndrome: Preliminary Acoustic Findings

Increased speaking rate is a commonly reported perceptual characteristic among males with fragile X syndrome (FXS). The objective of this preliminary study was to determine articulation rate-one component of perceived speaking rate-and vowel space characteristics of young males with FXS. Young males with FXS (n = 38), developmental age (DA)-matched males (n = 21), and chronological age (CA)-matched males (n = 16) were audiotaped while engaged in spontaneous conversation and a picture-naming task. Articulation rate in syllables per second during intelligible utterances and vowel space area/dispersion measures were acoustically determined for each speaker. Males with FXS did not articulate significantly faster than CA-matched males. Area and dispersion of the acoustic vowel space also were similar between the 2 groups. Males with FXS, however, used significantly shorter utterances and had a tendency to pause less often than CA-matched males. In addition, males with FXS exhibited greater intraspeaker variability of formants associated with the vowel /a/. These preliminary findings suggest that articulation rate may not be a primary factor contributing to perceived speaking rate of males with FXS. Limitations of the study relative to speech production tasks and utterance intelligibility are discussed.

Connected speech processes in developmental speech impairment: observations from an electropalatographic perspective

This paper uses a combination of perceptual and electropalatographic (EPG) analysis to explore the presence and characteristics of connected speech processes in the speech output of five older children with developmental speech impairments. Each of the children is shown to use some processes typical of normal speech production but also to use a range of more unusual processes at word junctures, with both inter- and intra-speaker variability apparent in the data. The findings suggest that the traditional focus on single word production in the assessment of developmental speech disorders needs to be modified to allow more extensive consideration of speech production in spontaneous speech.

Assessment of variation between and within speakers

While few individuals would argue that vocal cues can signal a person’s identity, it is difficult to specify exactly which parameter(s) provide the most salient information for speaker identification. Previous literature has suggested that speaking fundamental frequency, long-term spectra, vowel formant frequencies, and speech tempo can provide speaker-specific information. However, investigations focused on automatic speaker identification have provided less than satisfactory results. These findings could be related to how each acoustic parameter is measured or, more probably, to the idea that these acoustic parameters interact in specific ways that may be more obvious in the perceptual realm and may vary across speaking situations. To further complicate matters, individuals may speak more than one language or use multiple dialects. Little is known about the effect of code switching on voice production and identification. The purpose of this presentation is to present some of the relevant literature on voice recognition and factors related to misidentification. The role of intraspeaker variability will be discussed with a special emphasis on bilingualism and bidialectalism. Implications for voice production in augmentative and alternative communication devices will be described.

Categorical and gradient properties of assimilation in alveolar to velar sequences: evidence from EPG and EMA data

Place assimilation in English is now widely considered to be a gradual phonetic, not categorical process. This view is partly based on previous EPG evidence of partial alveolar assimilations which lack complete stop closure on the alveolar ridge but show a residual tongue blade/body gesture. This study reports EPG data from 10 speakers producing, at varying rates of speech, two experimental sequences, /n#k/ and /n#k/ (the latter a lexical velar–velar sequence with which apparent cases of complete assimilation can be compared). In fast speech, four distinct assimilatory strategies were identified. Two subjects never assimilated, four always assimilated in what appeared to be a complete fashion and the remaining four were the most interesting, showing considerable intra-speaker variability. Two of these four produced the expected continuum of assimilatory forms including partials. Unexpectedly, the other two produced either full alveolars or complete assimilations in the manner of a binary opposition. Follow-up EMA analysis yielded no evidence of the reduced coronal gestures found to be absent in the EPG-only data for two the speakers who, when they assimilated, did so in a complete fashion. Although no claims are made regarding higher-order representations, we interpret this as evidence of marked individual differences in assimilation strategy.

The intraspeaker variability of pitch in Mandarin and English subjects producing disyllabic words

Considering their use of pitch to convey lexical information, do tone language speakers develop greater control over pitch mechanisms, thus exhibiting less overall pitch variability than speakers of intonation languages? One previous researcher has argued that tone language speakers exhibit a ‘‘remarkably stable and precise absolute pitch template’’ when reading word lists [Deutsch et al., J. Acoust. Soc. Am. 106, 2267 (1999)]. The present study aims to investigate the intraspeaker pitch variability of Mandarin and English speakers when producing disyllabic words. Specifically, this research will examine (1) intraspeaker pitch variability between Mandarin and English speakers, (2) the degree of pitch stability exhibited by speakers of both languages, and (3) the effects of elicitation condition on a speaker’s pitch variability. Subjects will produce initially stressed, disyllabic words (similar segmental structure) in their native language elicited through picture identification. Randomized blocks containing similar target items will be presented under two elicitation conditions. To measure possible ‘‘list effects’’ the second elicitation condition will contain an intervening masking task. Measures of fundamental frequency change of target nuclei will be obtained at four locations: start point, end point, high point, and low point. A three-way ANOVA will be used to analyze the pitch variability data.

Linguistic databases Ed. by John Nerbonne (review)

BOOK NOTICES 209 Linguistic databases. Ed. by John Nerbonne . (CSLI lecture notes 77.) Stanford , CA: CSLI, 1998. Pp. xxi, 243. The papers in this collection were originally presented at the 'Linguistic Databases' conference, University of Groningen, 23-24 March, 1995. Because ofthe almostproverbial rapidity with which information technology develops, the collection as a whole is dated already, but there is still much of interest to be found. Not all papers read at the conference are in this volume, but the papers cover a wide range of subjects, mostly practical in nature, not theoretical. After a clear and readable introduction by Nerbonne , the papers are presented in no particularorder, though the editor groups the papers in five main areas: syntactic corpora and databases, phonetic databases , applications in linguistic theory, applications, and extending basic technologies. The papers themselves are not presented according to this grouping, however, and at first sight the book appears rather disorganized. The wide variety of subjects can be deduced from the titles of the papers presented: 'Test suites for natural language processing', 'From annotated corpora to databases: The SgmlQL language', 'Markup of a test suite with SGML', 'An open systems approach for an acoustic-phonetic continuous speech database: The S_tools database-management system ', "The reading database of syllable structure', ? database application for the generation of phonetic atlas maps', 'Swiss French polyphone and polyvar: Telephone speech databases to model inter- and intra-speaker variability', 'Investigating argument structure: The Russian nominalization database', "The use of a psycholinguistic database in the simplification of text for aphasie readers', "The computer learner corpus: A testbed for electronic EFL tools', 'Linking WordNet to a corpus query system', 'Multilingual data processing in the CELLAR environment ' . The issue whether to use open free systems or closed proprietary systems is addressed in several papers. Some papers present applications developed both in open and closed systems. This is one area where developments have been going very fast, and nowadays freely available databases are often as capable as their commercial counterparts. Some of the applications presented in this collection are available from the Internet, and url's are often given. The collection can serve as a good introduction to the field for relative outsiders as ample references and links are given. The papers themselves vary greatly in subject matter so not all will be of interest to every reader. My particular favorite was 'From annotated corpora to databases: the SgmlQL language '. [BOUDEWUN REMPT.j Understanding phonology. By Carlos Gussenhoven and Haike Jacobs. (Understanding language series.) London: Arnold, 1998. Pp. xii, 286. This textbook is intended as an introduction to phonology aimed at 'students with little or no prior knowledge of linguistics' (back cover). As in many other textbooks, it uses exercises as a learning tool. Two types ofexercises are proposed. The ones identified by a key, 'intended as an expository aid' (xi), are provided with a solution in an appendix (though it is not always so much a clear cut answer as a guide for reflection, which is, to my view, a lot better). The ones identified by a dot are intended as practice material, and no solution is offered. I thought the idea of having two types of exercises a good one since it gives the reader the opportunity both for individual work and for discussion with others. Also, whenever it may apply, an optimality theoretic analysis is offered to describe a phonological process. Ch. 1, "The production of speech', is a basic introduction to phonology, phonetics, and phonation. Ch. 2,'Some typology: Sameness and difference', cleverly covers the universal and language specific aspects of phonological structures and typology. Ch. 3,'Making the form fit', addresses phonological grammar and adaptation by presenting the nativization of loan words in both the rules and the constraints approaches. Ch. 4, 'Underlying and surface representations', Ch. 5, 'Distinctive features', and Ch. 6, 'Ordered rules', deal with the basic notions of generative phonology within the SPE type formalism and introduce the reader to the school of linear phonology. Ch. 7, ? case study: The diminutive suffix in Dutch', shows how these notions are applied. In Ch. 8, 'Levels of representation', Gussenhoven and Jacobs present an intermediate level of representation between the underlying representation and...

Acoustic variability in dysarthria and computer speech recognition

This paper examines the nature and extent of intra-speaker variability in normal and dysarthric speech and considers its relationship to computer speech recognition. Moderately dysarthric speech was found to exhibit greater intra-speaker variability on parameters of voice onset time, vowel duration, fricative duration and on vowel formants. Also noted for these speakers were patterns of merging acoustic boundaries, overlap in values for acoustic minimal pairs and shifting of the minimal pair boundary. Mildly dysarthric speech was found to follow a similar pattern to the controls with some additional shifts in contrast boundaries. Computer speech recognition levels were lower for moderately dysarthric speakers than for their associated controls. The mildly dysarthric speakers recognition pattern was similar to the controls. Correlation analysis of the relationship between acoustic variability and computer recognition found only a small proportion of the tokens to be significantly correlated at the 0.05 level of significance. Violation of minimal pair marking rules, along with merging of acoustic space and shifts in boundary positions between minimal pairs, were found to interact with variability to impact on computer recognition. The clinical implications are discussed with reference to the use of computer recognition as an augmentative communication tool.

Integrating pitch and LPC-residual information with LPC-cepstrum for text-independent speaker recognition.

In the speaker recognition, when the cepstral coefficients are calculated from the LPC analysis parameters, the prediction error, or LPC residual signal, is usually ignored. However, there is an evidence that it contains a speaker specific information. The fundamental frequency of the speech signal or the pitch, which is usually extracted from the LPC residual, has been used for speaker recognition purposes, but because of the high intraspeaker variability of the pitch it is also often ignored. This paper describes our approach to integrating the pitch and LPC-residual with the LPC-cepstrum in a Gaussian Mixture Model (GMM) based speaker recognition system. The pitch and/or LPC-residual are considered as an additional features to the main LPC derived cepstral coefficients and are represented as a logarithm of the F0 and as a filter bank mel frequency cepstral (MFCC) vector respectively. The second task of this research was to verify whether the correlation between the different information sources is useful for the speaker recognition task. For the experiments we used the NTT database consisting of high quality speech samples. The speaker recognition system was evaluated in three modes-integrating only pitch or only LPC-residual and integrating both of them. The results showed that adding the pitch gives significant improvement only when the correlation between the pitch and cepstral coefficients is used. Adding only LPC-residual also gives significant improvement, but in contrast to the pitch, using the correlation with the cepstral coefficients does not have big effect. The best results we achieved using both the pitch and LPC-residual and are 98.5% speaker identification rate and 0.21% speaker verification equal error rate compared to 97.0% and 1.07% of the baseline system respectively.

Analysis of ‘‘cold-affected’’ speech for inclusion in speaker recognition systems.

Parameters of speech produced during winter weather are analyzed to determine the potential influence of distorted voices due to common cold on automatic speaker recognition systems. This research examines speech recorded before, during, and after bouts with the common cold. ‘‘Cold-affected’’ speech shows noticeable acoustic and phonetic differences between test sessions. Examining the different states of one male speaker’s voice allows an opportunity to address the speaker intersession problem that exists in speaker recognition technology. Specific emphasis is placed on the contrasts in various sessions of TIMIT inspired sentence ‘‘She had your dark suit in greasy wash water all year.’’ It also analyzes phonetic contrasts and looks at differences in formant patterns. Phonetic transcriptions of ‘‘cold’’ and ‘‘normal’’ sessions reflect changes in place of articulation. Perceptual and acoustic analyses reveal pauses and epenthetic syllables that are not constant throughout all sessions. ‘‘Cold-speech’’ also produces noise caused by hoarseness and coughing. Research presented here shows evidence of variability beyond ‘‘normal’’ speech patterns. Contributing to the information on intraspeaker variability, this work provides a preliminary analysis of ‘‘cold-affected’’ speech in an effort to include ‘‘abnormal’’ speech and speech affected by communications disorders into the framework of automatic speaker recognition.

Inter- and intraspeaker variability in fundamental frequency of Thai tones

A measure was obtained of variability in fundamental frequency ( F 0) in citation forms of lexical tones. The language selected for investigation was Thai, a tone language with five lexical tones: mid, low, falling, high and rising. Twenty speakers participited in the experiment: 10 “young” male speakers and 10 “old” speakers, 5 male and 5 female. High-quality tape recordings were obtained of each subject's productions of a minimal set of five monosyllabic words. F 0 contours were extracted by a cepstral analysis. A comparison was made of inter- and intraspeaker variability in the production of the five Thai tones. Results of analysis of variance indicated that the degree of intersubject variability in F 0 was greater than intraspeaker across all five tones, that young and old speakers exhibited the same pattern of variability, and that variability in tone production differed depending on the lexical tone. The falling and rising tones exhibited smaller degrees of variability than the mid, low or high. Findings are interpreted to highlight the nature of F 0 variability, the relationsip of F 0 variability to amount of F 0 movement, and crosslinguistic differences in F 0 variability as a function of prosodic structure.

Intraproduction variability in jitter measures from elderly speakers

This study examined intraproduction variability in jitter measures from elderly speakers' sustained vowel productions and tried to determine whether mean jitter levels (percent) and intraspeaker variability on jitter measures are affected significantly by the segment of the vowel selected for measurement. Twenty-eight healthy elderly men (mean age 75.6 years) and women (mean age 72.0 years) were tape recorded producing 25 repeat trials of the vowels /i/, /a/, and /u/, as steadily as possible. Jitter was analyzed from two segments of each vowel production: (a) the initial 100 cycles after 1 s of phonation, and (b) 100 cycles from the most stable-appearing portion of the production. Results indicated that the measurement point selected for jitter analysis was a significant factor both in the mean jitter level obtained and in the variability of jitter observed across repeat productions. This study examined intraproduction variability in jitter measures from elderly speakers' sustained vowel productions and tried to determine whether mean jitter levels (percent) and intraspeaker variability on jitter measures are affected significantly by the segment of the vowel selected for measurement. Twenty-eight healthy elderly men (mean age 75.6 years) and women (mean age 72.0 years) were tape recorded producing 25 repeat trials of the vowels /i/, /a/, and /u/, as steadily as possible. Jitter was analyzed from two segments of each vowel production: (a) the initial 100 cycles after 1 s of phonation, and (b) 100 cycles from the most stable-appearing portion of the production. Results indicated that the measurement point selected for jitter analysis was a significant factor both in the mean jitter level obtained and in the variability of jitter observed across repeat productions.

Intra-speaker variability of the long term speech spectrum

400-channels constant bandwidth (12.5 Hz) Long Term Spectra (LTS) delivered by a BK 2033 analyser have been drawn from French utterances. The cross-correlation coefficient was used to investigate LTS residual intra-speaker variability both in inter- and intra-text conditions. Significant subject-dependent differences habe been revealed in both conditions. They indicate voice coherence variability in the speakers. Correlations lowering in inter-text condition moreover reveals the content-dependent nature of LTS. The need for basic LTS research prior to further application is therefore emphasized.

Intraspeaker variability in fundamental frequency stability: an age-related phenomenon?

The purpose of this investigation was to gather information on the extent to which intraspeaker variability on measures of jitter (%) and fundamental frequency standard deviation (F0 s.d.) is age related in women. Fifteen repeat productions of the vowels /i/, /a/, and /u/ from 22 young women (18-22 years) were analyzed for F0 s.d. and jitter. Findings for these young speakers were compared with those for elderly speakers tested previously (Linville and Korabic, 1987). Results indicate that the aging process brings about increases in the variability individual women demonstrate on measures of F0 stability when producing sustained vowels as steadily as possible. Further, young speakers differed markedly from elderly speakers in the pattern of frequency instability variations observed across the three vowels tested.