Vowel Tests Research Articles

Brainstem encoding of frequency-modulated sweeps is relevant to Mandarin concurrent-vowels identification for normal-hearing and hearing-impaired listeners

F0 contours convey the primary information of lexical Tones for Mandarin Chinese, and the processing of time-varying F0 contours is important for Mandarin concurrent-vowels identification (MCVI). In this work, we examined the relationship between frequency modulation (FM) detection of auditory system and MCVI in both normal-hearing (NH) and hearing-impaired (HI) listeners. Three experiments were conducted with the same subjects to measure their MCVI, FM detection limen (FMDL), and frequency following responses (FFRs) evoked by FM sweeps, respectively. To ensure that F0 contour was the primary cue utilized, mean F0s and durations were equalized among all test vowels in the MCVI experiment. To simulate the pattern of F0 contours of Mandarin vowels, linearly FM sweeps were used as stimuli in the FMDL and FFRs experiments. The results confirmed that the performance of HI listeners was significantly worse than that of NH listeners in all of the three measurements. Besides, FFRs evoked by FM sweeps had significantly lower tracking accuracy than those evoked by steady tones only for HI listeners. The correlation analysis further revealed that any two of the three measured indices were significantly correlated when the effects of age and absolute threshold were partialed out (|r| ≥ 0.502, p ≤ 0.017). These results suggested an association between the poor performance of HI listeners in the MCVI task and their degraded auditory function on FM detection, and such a behavioral degradation has emerged in the phase-locking activity at the brainstem level.

Neighborhood-conditioned coarticulation effects in French listener-directed speech

Words from dense phonological neighborhoods (Hi-ND words) are realized with greater vowel hyperarticulation and increased coarticulation in English, relative to words from sparser neighborhoods (Lo-ND words) [e.g., Wright 2004]. Here, the relation between coarticulation and neighborhood density is investigated for French by looking at patterns for anticipatory and carryover nasal coarticulation, coronal and uvular CV coarticulation, and V-to-C rounding coarticulation. Twelve native French speakers (northern France) produced 82 disyllabic words of French containing a context for one of these 5 types of coarticulation in phonetically similar high-low ND pairs. Test words were produced to a real listener who had to transcribe them in the instructed positions on a grid. F1, F2, and A1-P0 (a spectral measure of nasality) were measured at 5 timepoints across each test vowel. In addition to having more hyperarticulated (peripheral) vowel midpoints, Hi-ND words showed greater anticipatory nasality and stronger coronal transitions in appropriate contexts than Lo-ND words. Carryover nasality and uvular and rounding coarticulation did not differ across NDs. These results can be interpreted in light of the relation between coarticulation and contrast and the role these effects may serve in making Hi-ND words, which are subject to greater lexical competition, easier to perceive.

Speech perception adjusts to stable spectrotemporal properties of the listening environment

When perceiving speech, listeners compensate for reverberation and stable spectral peaks in the speech signal. Despite natural listening conditions usually adding both reverberation and spectral coloration, these processes have only been studied separately. Reverberation smears spectral peaks across time, which is predicted to increase listeners' compensation for these peaks. This prediction was tested using sentences presented with or without a simulated reverberant sound field. All sentences had a stable spectral peak (added by amplifying frequencies matching the second formant frequency [F2] in the target vowel) before a test vowel varying from /i/ to /u/ in F2 and spectral envelope (tilt). In Experiment 1, listeners demonstrated increased compensation (larger decrease in F2 weights and larger increase in spectral tilt weights for identifying the target vowel) in reverberant speech than in nonreverberant speech. In Experiment 2, increased compensation was shown not to be due to reverberation tails. In Experiment 3, adding a pure tone to nonreverberant speech at the target vowel's F2 frequency increased compensation, revealing that these effects are not specific to reverberation. Results suggest that perceptual adjustment to stable spectral peaks in the listening environment is not affected by their source or cause.

한국인 학습자의 영어 모음 인지: 새로운 L2 모음 범주와 비슷한 L2 모음 범주의 비교

The purpose of this study is to investigate how Korean learners perceive English vowels and further to test SLM which claims that new L2 vowel categories are more easily acquired than similar L2 vowel categories. Twenty Korean learners participated in English-to-Korean mapping test and English vowel identification test with target vowels /i, ɪ, u, ʊ, ɛ, ae/. The result revealed that Korean participants mapped the English pairs /i/-/ɪ/ and /u/-/ʊ/ onto single Korean vowel /i/ and /u/, respectively. in addition, both of English /ɛ/ and /ae/ were simultaneously mapped onto Korean /e/ and /ɛ/. This indicated that the Korean participants seemed to have perceptual difficulty for the pairs /i-ɪ/, /u-ʊ/, and /ɛ-ae/. The result of the forced-choice identification test showed that the accuracy of /ɪ, ʊ, ae/(ɪ: 81.3%, ʊ: 62.5%, ae: 60.0%) was significantly higher than that of /i, u, ɛ/(i: 28,8%, u: 28.8%, ɛ: 32.4%). Thus, the claim of SLM is confirmed given that /ɪ, ʊ, ae/ are new vowel categories whereas /i, u, ɛ/ are similar vowel categories. Further, the conspicuously low accuracy of the similar L2 vowel categories /i, u, ɛ/ was accounted for by over-generalization whereby the Korean participants excessively replaced L2 similar /i, u, ɛ/ with L2 new /ɪ, ʊ, ae/ as the participants were learning the L2 new vowel categories in the process of acquisition. Based on the findings this study, pedagogical suggestions are provided.

Distributional vowel training is less effective for adults than for infants. A study using the mismatch response.

Distributional learning of speech sounds (i.e., learning from simple exposure to frequency distributions of speech sounds in the environment) has been observed in the lab repeatedly in both infants and adults. The current study is the first attempt to examine whether the capacity for using the mechanism is different in adults than in infants. To this end, a previous event-related potential study that had shown distributional learning of the English vowel contrast /æ/∼/ε/ in 2-to-3-month old Dutch infants was repeated with Dutch adults. Specifically, the adults were exposed to either a bimodal distribution that suggested the existence of the two vowels (as appropriate in English), or to a unimodal distribution that did not (as appropriate in Dutch). After exposure the participants were tested on their discrimination of a representative [æ] and a representative [ε], in an oddball paradigm for measuring mismatch responses (MMRs). Bimodally trained adults did not have a significantly larger MMR amplitude, and hence did not show significantly better neural discrimination of the test vowels, than unimodally trained adults. A direct comparison between the normalized MMR amplitudes of the adults with those of the previously tested infants showed that within a reasonable range of normalization parameters, the bimodal advantage is reliably smaller in adults than in infants, indicating that distributional learning is a weaker mechanism for learning speech sounds in adults (if it exists in that group at all) than in infants.

Perception and production of English vowels by Korean learners: A Case study

The main objective of this study was to investigate the relationship between perception and production of English vowels by Korean EFL learners. To this end, 21 Korean students participated in perception and production tests in order to probe the extent to which the perception and production accuracies conform to each other. Overall, the participants had difficulty perceiving and producing English vowels since the mean accuracies of perception and production only amounted to 60%. The results showed that perception accuracy was not always higher than production accuracy, which runs counter to the traditional assumption that perception precedes production. Moreover, it was revealed that no significant correlation between perception and production was found, thus suggesting that perception and production may not be related and that perception and production capabilities may be developing independently depending on vowels and individuals. As for the results of the perception test, monophthong and short vowels were more difficult to perceive in relation to their counterpart diphthong and long vowels. Regarding the results of the production test, the participants were better at producing tense vowels than lax vowels. In addition, confusion matrixes for perception and production were presented and confusion patterns as well as the direction of confusability are discussed in detail.

Estimation of the Voice Source from Speech Pressure Signals: Evaluation of an Inverse Filtering Technique Using Physical Modelling of Voice Production

Objective: The goal of the study is to use physical modelling of voice production to assess the performance of an inverse filtering method in estimating the glottal flow from acoustic speech pressure signals. Methods: An automatic inverse filtering method is presented, and speech pressure signals are generated using physical modelling of voice production so as to obtain test vowels with a known shape of the glottal excitation waveform. The speech sounds produced consist of 4 different vowels, each with 10 different values of the fundamental frequency. Both the original glottal flows given by physical modelling and their estimates computed by inverse filtering were parametrised with two robust voice source parameters: the normalized amplitude quotient and the difference (in decibels) between the levels of the first and second harmonics. Results:The results show that for both extracted parameters the error introduced by inverse filtering was, in general, small. The effect of the distortion caused by inverse filtering on the parameter values was clearly smaller than the change in the corresponding parameters when the phonation type was altered. The distortion was largest for high-pitched vowels with the lowest value of the first formant. Conclusions: The study shows that the proposed inverse filtering technique combined with the extracted parameters constitutes a voice source analysis tool that is able to measure the voice source dynamics automatically with satisfactory accuracy.

Pre-attentive detection of vowel contrasts utilizes both phonetic and auditory memory representations

Event-related brain potentials (ERP) were recorded to infrequent changes of a synthesized vowel (standard) to another vowel (deviant) in speakers of Hungarian and Finnish language, which are remotely related to each other with rather similar vowel systems. Both language groups were presented with identical stimuli. One standard-deviant pair represented an across-vowel category contrast in Hungarian, but a within-category contrast in Finnish, with the other pair having the reversed role in the two languages. Both within- and across-category contrasts elicited the mismatch negativity (MMN) ERP component in the native speakers of either language. The MMN amplitude was larger in across- than within-category contrasts in both language groups. These results suggest that the pre-attentive change-detection process generating the MMN utilized both auditory (sensory) and phonetic (categorical) representations of the test vowels.

The discovery of natural classes by non-native listeners

Using natural disyllabic stimuli, English speakers were trained to sort one, two, or three pairs of German nonlow rounded vowels into two cat-egories, then a new pair was added and their ability to sort it into the same categories was tested. Each pair contrasted in [high], [tense], and [back]. Category membership was determined by only one of these features. Three-pair training provided enough information to tell which one; two-pair narrowed it down to two possibilities; one-pair left it open. Listeners in three experiments using different pairs of test vowels sorted the test vowels significantly better as the number of training pairs increased. This suggests they were inducing feature-based natural classes. The rival exemplar model of Nosofsky [J. Exp. Psych.: Gen. 115, 39–57 (1986)] would say that while more pairs bring more phonetic variety, they also exemplify more robustly the dimensions along which the categories differ phonetically, and thereby shift attentional weight to those dimensions. Similar experiments will pit these two hypotheses against one another with training pairs which contrast only in the feature that defines the sort categories. If listeners are using feature-based natural classes, their performance should not improve with more training pairs; if they are using exemplars, it should. [Work supported by NIH.]

Pre-recognition measures of speaking rate

The accuracy of speech recognition systems is known to be affected by fast speech. If fast speech can be detected by means of a measure of speaking rate, the acoustic as well as language models of a speech recognition system can be adapted to compensate for fast speech effects. We have studied several measures of speaking rate which have the advantage that they can be computed prior to speech recognition. The proposed measures have been compared with conventional measures, viz., word and phone rate on the TIMIT database. Some of the proposed measures have significant correlations with phone rate and vowel duration. We have shown that the mismatch between actual and expected durations of test vowels reduces if the vowel duration models are adapted to speaking rate, as estimated by the proposed measures. These measures can be computed from features commonly employed in speech recognition, do not entail significant additional computational load and do not need labeling or segmentation of unknown utterance in terms of linguistic units.

Perceptual and acoustic aspects of lexical vowel reduction, a sound change in progress

In the present study 20 Dutch male speakers were asked to read aloud 47 test words in a word list and in short sentences. Part of this word set was also named by them through the presentation of pictures. A group of 20 listeners was asked to identify an unstressed vowel in all of these test words. The vowel responses of listeners were recoded into two broad categories: “full vowel” and “schwa”. Our aims were (1) to find out to what extent listeners are able to unambiguously distinguish between these two categories, (2) to investigate the influence of the frequency of occurrence of words on the classification of the test vowels, (3) to investigate the influence of speaking styles on the classification of the test vowels by comparing the speech conditions “word list”, “pictures” and “sentences”. The experimental results showed that (1) listeners often could not unambiguously classify the test vowels, especially if these occurred in interstress position, (2) the number of schwa responses was much higher for vowels in words with a relatively high frequency of occurrence, (3) the number of schwa responses increased in a more casual speaking style. Acoustic measurements on the test vowels revealed a clear relation between the perceptual results and the acoustic features of the vowels. Although the preconditions for the sound change “full vowel → schwa” in several Dutch words are excellent, the actual completion of the sound change is in our view to a large extent blocked by the rather close correspondence between Dutch vowel sounds and their orthographic representations.

Speech audiometric trial using synthetic vowels produced with DECtalk

A speech audiometric trial is described for testing the degree of hearing loss and its pathogenesis. Test vowels and distorted speech sounds were synthesized with DECtalk for use in speech audiometric tests. Results showed that 100% articulation can be obtained with undistorted synthesized vowels, while reasonable articulation scores can be obtained with distorted synthetic speech signals. This method could solve some of the problems inherent in use of recorded natural Japanese speech sounds as acoustic stimuli during otological examinations.

A code in the nodes

Four two-layer associative networks (one layer of hidden nodes) were trained to classify spectra extracted from naturally produced vowels of one talker, using the hack-propagation learning algorithm [Rumelhart et al., Nature 323, 533–536 (1986)]. The inputs to the networks were either LPC spectra or “perceptual” spectra, obtained by convolving the Bark-transformed spectrum with an asymmetric filter. The networks trained on LPC input classified test vowels more accurately than the networks trained on the perceptually based spectra. The networks' solutions were examined to determine (a) whether it was possible to characterize the features underlying the solutions and (b) whether the input spectral representation affected the nature of the solutions. Principal components analyses of the activation patterns of the hidden units showed that three dimensions accounted for over 80% of the variance for all networks. For networks trained on “perceptual” spectra, the first three dimensions were highly correlated with F2/F1 ratio, F3/F2 ratio, and F3, respectively. For LPC input, the first three dimensions correlated with F1, F3/F2 ratio, and F2/F1 ratio. Thus these networks classified the vowels using features related to the formant space.

The identifiability of isolated whispered and phonated vowel samples

Two experiments were designed to study whispered and phonated vowel perception. In each experiment, the whispered and phonated test vowel samples were produced by the same subjects. In experiment 1, vowel samples produced by adult female subjects were examined. In experiment 2, vowel samples produced by adult male subjects were examined. In both experiments, whispered and phonated isolated vowels were identified correctly by individual judges in a majority of instances. Phonated vowel samples, generally, were identified correctly more often than whispered vowel samples. The vowels /i, æ, u/ were identified best among phonated vowels; the vowel /i/ was identified best among whispered vowels.

Formant-frequency differences between isolated whispered and phonated vowel samples produced by adult female subjects.

This study was designed to investigate the formant frequencies of phonated and whispered productions of five test vowels (/i/, /u/, /ae/, /a/). Each test vowel was sustained twice in isolation--once phonated, one whispered--by 20 adult female subjects. The phoneme represented by each recorded production was identified independently by 11 listeners. Only those samples identified by 6 or more of the listeners as the vowel intended were retained for a further (acoustic) analysis. An acoustic spectrum of each retained sample was obtained to permit formant measurements. To provide the clearest formant delineation possible in our lab, the phonated samples were analyzed by broadband spectrography and the whispered samples by very narrow-band spectrography. This report presents the formant frequencies (F1-F3) of the test productions as measured from the acoustic vowel spectra and the formant-frequency differences between the phonated and whispered productions. In general, the results showed a trend for whispered vowel formants to be higher in frequency than phonated vowel formants, but that trend was only strongly evident for F1.

An acoustic comparison of isolated whispered and phonated vowel samples produced by adult male subjects

This study was designed to investigate the formant frequencies and amplitudes of phonated and whispered productions of five test vowels (/i/, /u/, /æ/, /ɑ/, and /ʌ/). Each test vowel was sustained six times in isolation, three times in whisper and three times in phonation, by 15 adult male subjects. The phoneme represented by each recorded production was identified independently by 11 listeners. An acoustic spectrum of each sample was obtained through the use of a real-time narrow-band spectrum analyzer to permit formant measurements. This report presents the formant frequencies and amplitudes (F 1 to F 3 ) of the test productions as measured from the acoustic vowel spectra and the formant frequency and amplitude differences between the phonated and whispered productions. In general, the results showed a trend for whispered vowel formants to be higher in frequency than phonated vowel formants, but that trend was only strongly evident for F 1 .

Vowel Intrinsic Pitch in Connected Speech

Abstract We investigated whether ‘intrinsic pitch’ (IP) effects occur in connected speech. Fundamental frequency differences between German high and low vowels were measured in two different experimental conditions: (1) a typical laboratory task in which a carrier sentence served as a frame for test vowels; (2) a paragraph reading task in which test vowels occurred in a variety of prosodic environments. By comparing test vowels in comparable segmental and prosodic environments, it was shown that the IP effect does occur in connected speech, but that the size of the differences between high and low vowel fundamental frequency is somewhat smaller than in carrier sentences. Some tentative hypotheses are presented concerning the influence of prosodic environment on the magnitude of IP differences.

Identification of normal and abnormally rough vowels by spectral noise level measurements

Our goal in this study was to learn whether vowel spectral noise level (SNL) measurements might differentiate between normal and abnormally rough test vowel samples. The vowel samples studied were obtained originally for two previous investigations. In one study, 20 normal-speaking subjects phonated in isolation each of five test vowels (/u/, /i/, /λ/, /a/, /æ/) both normally and with simulated abnormal vocal roughness. In the other, 20 subjects who individually presented a laryngeal pathology phonated each of the same test vowels in isolation. The 300 test samples available from the two studies were obtained at one intensity and mouth-to-microphone distance (75 dB SPL at 15 cm). From the original studies, narrow-band (3-Hz) acoustic spectra of the vowel samples and SNL measurements were available. For this study, a mean of 25 noise level measurements per vowel sample served as an SNL index for each test production. Additionally, for this study, the test samples were individually classified dichotomously as normal or as abnormally rough in quality by a panel of nine listeners. Data analysis resulted in the identification of an SNL mean value (termed the SNL “criterion”) that appeared to differentiate effectively between the perceptually normal and abnormally rough samples of each test vowel. In general, the findings appeared to support the hypothesis that vowel SNL measurements could be used to identify normal and abnormally rough vowel samples.

The time course of vowel context effects

At previous meetings of the Society we presented evidence for the existence of two different perceptual mechanisms mediating anchor‐induced contrast effects found with vowels. Extra presentations of an [i] endpoint to subjects identifying an [i]‐[I] series caused changes in sensitivity. Anchoring with [I] produced criterion shifts which were dependent on the availability of auditory memory. Since [i] and [I] differ in tenseness, intrinsic duration in natural speech, and acoustic status (point versus nonpoint), the differences in perceptual effects of these vowels cannot be ascribed to any one factor. To explore the role of these variables in vowel context effects, a number of different vowel series from different parts of the vowel space have been examined using a successive contrast paradigm. In this procedure, a context vowel is presented first, followed at varying temporal intervals by a test vowel. Results obtained using this procedure provide new evidence concerning the perceptual mechanisms underlying vowel anchor effects. Further, the pattern of these results obtained with different vowel series have important implications for theories of vowel perception. [Work supported by NSF and NIMH.]

Vocal register effects on vowel spectral noise and roughness: Findings for adult males

This study was the second in a series designed to investigate the effects of vocal register (vocal, fry, modal, and falsetto) on the perceived roughness and spectral noise level of isolated test vowels. The first study (reported previously) was concerned with such effects on the phonations of adult females; in this study the phonations of adult males were investigated. Each of 15 male subjects produced at a controlled intensity each of two test vowels (/u/ and /æ/) in each of three vocal registers. Eleven listeners subsequently rated the test samples for roughness on a 5-point equal-appearing intervals scale. The criterion measure of roughness for each sample was the median of listener ratings (MRR). Each sample was also analyzed to produce its 3-Hz bandwidth acoustic spectrum from which measures of vowel spectral noise were obtained. The criterion measure of spectral noise level (SNL) for each test sample was the mean of 25 measures taken in the frequency range from 100 to 2600 Hz. The major finding was that the MRR and SNL for productions of both test vowels diminished significantly across vocal registers; i.e., from fry, to modal, to falsetto. In general, the present findings for males appeared consistent with those we reported earlier for phonations by adult females.