Vowel Center Research Articles

The Use of Static and Dynamic Cues for Vowel Identification by Children Wearing Hearing Aids or Cochlear Implants.

To examine vowel perception based on dynamic formant transition and/or static formant pattern cues in children with hearing loss while using their hearing aids or cochlear implants. We predicted that the sensorineural hearing loss would degrade formant transitions more than static formant patterns, and that shortening the duration of cues would cause more difficulty for vowel identification for these children than for their normal-hearing peers. A repeated-measures, between-group design was used. Children 4 to 9 years of age from a university hearing services clinic who were fit for hearing aids (13 children) or who wore cochlear implants (10 children) participated. Chronologically age-matched children with normal hearing served as controls (23 children). Stimuli included three naturally produced syllables (/ba/, /bi/, and /bu/), which were presented either in their entirety or segmented to isolate the formant transition or the vowel static formant center. The stimuli were presented to listeners via loudspeaker in the sound field. Aided participants wore their own devices and listened with their everyday settings. Participants chose the vowel presented by selecting from corresponding pictures on a computer screen. Children with hearing loss were less able to use shortened transition or shortened vowel centers to identify vowels as compared to their normal-hearing peers. Whole syllable and initial transition yielded better identification performance than the vowel center for /ɑ/, but not for /i/ or /u/. The children with hearing loss may require a longer time window than children with normal hearing to integrate vowel cues over time because of altered peripheral encoding in spectrotemporal domains. Clinical implications include cognizance of the importance of vowel perception when developing habilitative programs for children with hearing loss.

Mandarin tone identification with subsegmental cues in single vowels and isolated words

ABSTRACTThis study investigated the contributions of subsegmental cues [i.e., vowel center, vowel onset, and consonant (C)-vowel (V) transition] to Mandarin tone identification. In Experiment 1, single vowels were edited to preserve different proportions of vowel centers or vowel onsets. In Experiment 2, isolated Mandarin words were processed to produce two conditions of consonant plus C-V transition, and C-V transition. The synthesized stimuli were presented to normal-hearing native-Mandarin listeners for tone identification. Experimental results showed that, at the same proportion, vowel centers carried more information than vowel onsets for tone identification. The effect of vowel onsets on tone identification depended on the context of testing materials, with increased identification accuracy from single vowels to C-V transitions in isolated words. The duration of subsegmental cues could potentially predict the accuracy of subsegment-based tone identification.

Clear speech benefits for perception of center-only and edge-only syllables

Clear speech is a speaking style that has been shown to enhance intelligibility in noise; however, the underlying reasons for this enhancement are less well understood. One hypothesis is that listeners require less acoustic information to identify syllables spoken in clear than in conversational speech, allowing them to make use of briefer dips in fluctuating noise. The present study tests this hypothesis by presenting six /bVd/ syllables (“bead, bid, bed, bayed, bad,” and “bod”), produced in clear and conversational speech styles, to 20 monolingual native English-speaking listeners in a six-alternative forced-choice task. These syllables were modified to present varying portions of the syllable to listeners (20, 40, 60, or 80 ms of the syllable preserved). First, center-only syllables were created, in which acoustic information around the vowel midpoint was preserved. Second, edge-only syllables were created, in which information from the vowel center was silenced and formant transitions preserved, with ...

Evidence for a general auditory processing deficit in developmental dyslexia from a discrimination paradigm using speech versus nonspeech sounds matched in complexity.

It is unknown whether phonological deficits are the primary cause of developmental dyslexia or whether they represent a secondary symptom resulting from impairments in processing basic acoustic parameters of speech. This might be due, in part, to methodological difficulties. Our aim was to overcome two of these difficulties: the comparability of stimulus material and task in speech versus nonspeech conditions. In this study, the authors (a) assessed auditory processing of German vowel center stimuli, spectrally rotated versions of these stimuli, and bands of formants; (b) used the same task for linguistic and nonlinguistic conditions; and (c) varied systematically temporal and spectral parameters inherent in the German vowel system. Forty-two adolescents and adults with and without reading disabilities participated. Group differences were found for all linguistic and nonlinguistic conditions for both temporal and spectral parameters. Auditory deficits were identified in most but not all participants with dyslexia. These deficits were not restricted to speech stimuli-they were also found for nonspeech stimuli with equal and lower complexity compared with the vowel stimuli. Temporal deficits were not observed in isolation. These results support the existence of a general auditory processing impairment in developmental dyslexia.

Perceptual contribution of vowel sub-segments to Mandarin tone identification

Recent noise-replacement studies showed that (1) vowels carried more intelligibility information in Mandarin sentence recognition, and (2) a little vowel onset portion could significantly increase the intelligibility when it was added to the consonant-only Mandarin sentences. This study further evaluated the perceptual contribution of vowel sub-segments to Mandarin tone identification. The original duration-normalized vowels (FULL) were modified to produce two types of stimulus, i.e., (1) Left-only [LO (p)], which preserved p = 10% to 50% of the initial vowel portion, and replaced the rest vowel portion with speech-shaped noise (SSN), and (2) center-only [CO (p)], which preserved p = 15% to 60% of the center vowel portion, and replaced the rest initial and final vowel portions with SSN. Tone identification scores were collected from 20 normal-hearing native-Mandarin listeners. Results in the present study showed that (1) Mandarin tone perception at the LO (10%) condition was slightly higher than the chance level (i.e., 25%); (2) tone identification at the CO (60%) condition was not significantly different with that at the FULL condition. These findings suggest that vowel onset portion provides information redundant to vowel centers for Mandarin tone identification, and vowel centers contain sufficient information for reliable Mandarin tone identification.

Vowel identification by cochlear implant users: Contributions of static and dynamic spectral cues

Previous research has shown that normal hearing listeners can identify vowels in syllables on the basis of either quasi-static or dynamic spectral cues; however, it is not known how well cochlear implant (CI) users with current-generation devices can make use of these cues. The present study assessed vowel identification in adult CI users and a comparison group of young normal hearing (YNH) listeners. Stimuli were naturally spoken /dVd/ syllables and modified syllables that retained only quasi-static spectral cues from an 80-ms segment of the vowel center ("C80" stimuli) or dynamic spectral cues from two 20-ms segments of the vowel edges ("E20" stimuli). YNH listeners exhibited near-perfect performance for the unmodified (99.8%) and C80 (92.9%) stimuli and maintained good performance for the E20 stimuli (70.2%). CI users exhibited poorer average performance than YNH listeners for the Full stimuli (72.3%) and proportionally larger reductions in performance for the C80 stimuli (41.8%) and E20 stimuli (29.0%). Findings suggest that CI users have difficulty identifying vowels on the basis of spectral cues in the absence of duration cues, and have limited access to brief dynamic spectral cues. Error analyses suggest that CI users may rely strongly on vowel duration cues when those cues are available.

Assessing the perceptual contributions of vowels and consonants to Mandarin sentence intelligibility

This study investigated the perceptual contributions of vowels and consonants to Mandarin sentence intelligibility. Mandarin sentences were edited using a noise-replacement paradigm to preserve various amounts of segmental information and presented to normal-hearing listeners to recognize. The vowel-only Mandarin sentences yielded a remarkable 3:1 intelligibility advantage over the consonant-only sentences. This advantage is larger than that obtained with English sentences, suggesting that vowels may have a greater contribution to sentence intelligibility in Mandarin than in English. Although providing information redundant to contributions from vowel centers, a little vowel-consonant boundary transition would significantly improve the intelligibility of the consonant-only Mandarin sentences.

Contributions of static and dynamic spectral cues to vowel identification by cochlear implant users

Relatively little is known about cochlear implant (CI) users' ability to make use of static versus dynamic spectral cues in vowel perception tasks. The present study measured vowel identification in CI users and young normal hearing (YNH) listeners using naturally produced /dVd/ stimuli (deed, did, Dade, dead, dad, dud, and Dodd). Vowel identifcation was tested for (1) the unmodified syllables, (2) syllables modified to retain only 60 or 80 ms of the vowel center (center-only conditions), and (3) syllables modified to retain only 30 or 40 ms of the initial and final vowel transitions, with vowel duration neutralized (edges-only conditions). YNH listeners achieved high levels of performance for the unmodified stimuli (avg. 99.8%) and for the center-only stimuli (90.8%); their performance dropped to more moderate levels (68.1%) for the edges-only stimuli. CI users demonstrated moderate performance for the unmodified stimuli (avg. 72.0%) but demonstrated substantially poorer performance for both the center-only (41.1%) and edges-only stimuli (27.8%). Findings suggest that CI users (1) have difficulty identifying vowels in syllables when one or more cues are absent and (2) rely more strongly on quasi-static cues from the vowel center as compared to dynamic cues at the syllable edges.

Toronto English vowels: Deriving the vowel space from conversational data.

This presentation develops an overview of the Toronto English vowel space using data derived from the unscripted, casual speech typical of sociolinguistic interviews, rather than the more “formal” avenue of laboratory speech. Drawing on a corpus of sociolinguistic interviews collected by Walker & Hoffman (“language contact, linguistic variation and ethnic identity in Toronto English” SSHRC SRG 410‐2008‐2048) and using multiple‐point formant measurements of stressed vowels in “clean” environments (e.g., avoiding coda nasals and liquids), this study develops views of the vowel space that illustrate modal or average vowel centers, within‐category scatter, between‐category dispersion, and VISC. In refining this kind of vowel data collection and analysis, this research provides a baseline for further studies, such as to describe vowel differences in “ethnically” marked varieties of English, to characterize adjustments made in specific phonological contexts, to identify previously undescribed phonetic variation in Canadian English, and to compare similar data from other varieties of English.

Vowel identification by younger and older listeners: Relative effectiveness of vowel edges and vowel centers

Young normal-hearing (YNH) and older normal-hearing (ONH) listeners identified vowels in naturally produced /bVb/ syllables and in modified syllables that consisted of variable portions of the vowel edges (silent-center [SC] stimuli) or vowel center (center-only [CO] stimuli). Listeners achieved high levels of performance for all but the shortest stimuli, indicating that they were able to access vowel cues throughout the syllable. ONH listeners performed similarly to YNH listeners for most stimuli, but performed more poorly for the shortest CO stimuli. SC and CO stimuli were equally effective in supporting vowel identification except when acoustic information was limited to 20 ms.

Identification of transition‐only and steady‐state vowels by young and older normal‐hearing listeners.

Normal‐hearing listeners can identify vowels on the basis of either dynamic or steady‐state (SS) cues. To determine which cues are more effective, vowel identification was measured for full, transition‐only (TN) and SS versions of naturally produced exemplars of the syllables “beeb, bib, babe, beb, bab, and bob.” TN stimuli retained 10, 20, 30, 40, 60, or 80 ms of the consonant‐vowel and vowel‐consonant transitions and were neutralized in overall duration. SS stimuli retained 10, 20, 30, 40, 60, or 80 ms of the vowel center but eliminated transitions. Young normal‐hearing (YNH) and older normal‐hearing (ONH) listeners were assessed. Performance declined as the duration of acoustic information decreased. On average, decrements were similar for TN and SS stimuli except at the shortest duration (TN10 and SS20), where average performance was poorer for the TN stimulus. However, relative performance for short‐duration TN and SS stimuli varied substantially across vowels. TN and SS stimuli produced similar perf...

Proficient bilinguals require more information for vowel identification than monolinguals

Even proficient bilinguals have been shown to experience more difficulty understanding speech in noise than monolinguals. One potential explanation is that bilinguals require more information than monolinguals for phoneme identification. We tested this hypothesis by presenting gated, silent-center vowels to two groups of listeners: (1) monolingual American English speakers and (2) proficient Spanish–English bilinguals, who spoke unaccented or mildly accented English. To create the stimuli, two American English speakers were recorded as they read the following items: ‘‘beeb, bibb, babe, bebb, babb,’’ and ‘‘bob.’’ Duration-preserved silent-center versions of three tokens of each item were created by retaining varying amounts of the CV and VC transitions (10, 20, 30, or 40 ms) and attenuating the remainder of the vowel center to silence. Duration-neutral versions of silent-center tokens were created by lengthening or shortening the silent portion to match the tokens vowel duration to the average for all the tokens. Listeners identified the unedited (full vowel), duration-preserved, and duration neutral silent-center tokens in a six-alternative forced-choice task. The two groups of listeners identified the unedited tokens with similar accuracy. In the silent-center conditions, however, the bilinguals identified the stimuli less accurately than the monolinguals. [Work supported in part by NIH-NIDCD Grant No. 1R03DC005561-01A1.]

Dialect variation and formant frequency: The American English vowels revisited

Vowel production data collected from 15 southern Californian English-speaking monolinguals is compared with data reported by Hillenbrand et al. [J. Acoust. Soc. Am. 97, 3099–3111 (1995)] and Peterson and Barney [J. Acoust. Soc. Am. 24, 175–184 (1952)]. Recordings were made of nine women and six men producing multiple repetitions of [i, ɪ, e, ɛ, æ, u, ʊ, o, ɑ, ʌ, ɹˌ] in three consonant contexts. The frequencies of the first three formants were measured by simultaneous comparison of wideband spectrograms, narrow-band FFT spectral slices, and LPC spectra taken at vowel center, or steady state where available. The Southern Californian data are seen to differ greatly from that described by Peterson and Barney (1952) and Hillenbrand et al. (1995).

Vowel identification from harmonic contours of vowel centers using an automatic algorithm that eliminates windowing error.

Using successive nonoverlapping 22.5-ms windows, 16-kHz sampling, no filtering, an automatic algorithm locates five successive windows of minimum spectral velocity, describes a subwindow equal to some multiple of the natural period of F0, and maps the subwindow onto the unit circle, the interval 0, 2π. Consequently, the Fourier analysis is performed on a window where the signal is exactly periodic. Because the spectrum contains no extraneous numerical sidebands it is precise, and consists only of natural harmonics in the acoustic signal. The first 32-integer multiples of F0 are sufficient to describe the spectrum. J. D. Miller’s [J. Acoust. Soc. Am. 85, 2114–2134 (1989)] log F01/3 shift increases recognition of 12 vowels (men, women, and children) in the Hillenbrand et al. [J. Acoust. Soc. Am. 97, 3099–3111 (1995)] data set from 52% to 75% using a Euclidean classifier (EC—with jackknife). Cosine series (12) were used to compare our Betancourt spectrum (EC: 76%) with the Hamming window (EC: 61%). Quadratic discriminant function analysis + log F0 (79%) adds only 3% to our best EC result. With this spectrum and a log F01/3 shift, most vowel information is clearly captured by a simple EC.

Acoustic signal analysis using Gabor‐type windows.

Transients in speech and other acoustic signals are usually irregular and very difficult to analyze because of their short and nondeterministic duration. Therefore the traditional Fourier analysis technique does not give satisfying results. The classical Gabor technique uses a Gaussian waveform window. However, it is not numerically stable and it has slow convergence. In order to do time‐frequency analysis of acoustic signals, a window transform is proposed that is derived from a Gaussian using the Zak transform. The resulting window possesses good locality properties in both the time and frequency domains. The computational procedure, which based on the FFT algorithm, is robust and fast. It will be shown that transform yields accurate estimates of onset and offset times for periodic and aperiodic signals as well as for vowel centers and the ramp frequencies of formant transitions.

Word hypothesizer for continuous speech decoding using stress-vowel centered bidirectional tree searches

A word hypothesis module for speech decoding consists of four submodules: vowel center detection, bidirectional tree searches around each vowel center, forward-backward pruning, and additional short words hypotheses. By detecting the strong energy vowel centers, a vowel-centered lexicon tree can be placed at each vowel center and searches can be performed in both the left and right directions, where only simple phone models are used for fast acoustic match. A stage-wise forward-backward technique computes the word-beginning and word-ending likelihood scores over the generated half-word lattice for further pruning of the lattice. To avoid potential miss of short words with weak energy vowel centers, a lexicon tree is compiled for these words and tree searches are performed between each pair of adjacent vowel centers. The integration of the word hypothesizer with a top-down Viterbi beam search in continuous speech decoding provides two-pass decoding which significantly reduces computation time.

Some observations of women’s and children’s F2 frequencies: anatomical considerations and coarticulatory effects

One controversy concerning children’s speech is whether they coarticulate more, less, or about the same as adults. A confounding factor in studies addressing this issue is that vocal tract morphology varies with age and sex: Ratios of pharyngeal to oral cavity length are smallest for children, largest for men, and intermediate for women. Consequently, differences among men’s, women’s, and children’s F2s should be greater for front unrounded than for back unrounded vowels [G. Fant, SpeechSoundsandFeatures (1973)]. Pharyngeal and oral cavity lengths and F2s for /i/ and /ɑ / at vowel center of CV syllables were compared for women and 3-year-olds, with men serving as a reference. Results showed that children’s F2s did not differ from those of women as much as would be predicted based on anatomical considerations. This finding was used to evaluate the magnitude of children’s anticipatory coarticulation. It was found that F2s of children’s /i/ and /ɑ / differed more from those of women at voicing onset than at vowel center, suggesting that children were coarticulating more. Two other findings supported this conclusion: (1) Ratios of F2i to F2ɑ were more similar for children at voicing onset and at vowel center than they were for women at these two locations; and (2) ratios of F2 at voicing onset to F2 at vowel center for a given vowel were closer to 1.00 for children than for women. [Work supported by NIH.]

Phonetic underspecification in Marshallese.

The vowels of Marshallese have been analyzed as contrasting only along the height dimension, variation along the front–back dimension being predictable by the secondary articulations of adjacent consonants. Spectrographic evidence suggests this analysis is correct and that the surface allophones can be derived by interpolating between C1 and C2 targets across a vowel that is unspecified for F2. To test this hypothesis, /C1VC2/ words (Cn={/pj/, /pveeswirl/, /tj/, /tveeswirl/} and V={/schwa/, /barred eye/, /a/}) were recorded from four speakers. Then F2 was measured from vowel onset to offset. Preliminary findings based on two of the speakers shows that while F2 variation at the vowel center is statistically significant as a function of vowel category, the proportion of variance accounted for by vowel category is very low compared to that accounted for by the secondary articulation of C1 and C2. This is interpreted as evidence that a vocalic F2 target may not be necessary in modeling the trajectory. Multiple linear regression tests support this interpretation, showing that the vowel midpoint can be accurately predicted without reference to a vowel target. Regression analysis was performed on 15 time-normalized points with and without an intercept k, where k represents an independent vowel contribution to F2. The results show that the contribution of k is negligible (±0.001). Data from the two remaining speakers will be incorporated and discussed in terms of phonetic underspecification in a target-interpolation model.

Difference limens for synthetic vowel spectra

Difference limens (DLs) along synthetic vowel continua were measured as distances in the auditory-perceptual space proposed by J. D. Miller [J. Acoust. Soc. Am. 85, 2114–2134 (1989)]. The continua represent straight lines that lie parallel with one of the axes of the space (x′, y′, and z′). Groups of six continua share a common center value, or reference point. Movement along these continua result in distinct patterns of frequency change for F1, F2, and F3 relative to the reference point formant values. These patterns vary with the direction and axis of movement. Reference points were selected from the interiors of ten vowel zones and seven boundary areas between vowel zones. An adaptive up—down procedure employing a cued, two-alternative, forced-choice (2AFC) task was utilized to estimate the 79.5% correct point along each continuum. Vowel tokens representing the reference points served as the cue stimuli. DLs were estimated twice for each continuum from each of four subjects. In general, the results of this experiment reflect smaller DLs for vowel formant frequencies than have been found in the past and suggest that while an overall average DL for distance in the auditory-perceptual space may be estimated, DLs vary significantly with the axis of movement. Additionally, no significant difference was found between continua associated with vowel centers and vowel boundaries. These results will be compared with similar results estimating DLs for single-formant variation in vowels. [Work supported by NIDCD.]

Context-modulated discrimination of similar vowels using second-order connectionist networks

Discrimination of medial adjacent vowels in the context of voiced and unvoiced stop consonants using connectionist networks is investigated. Separate discrimination networks were generated for one speaker from samples of the vowel centers of [e,æ] for the six contexts [b,d,g,p,t,k]. A single context-independent network was similarly generated. The context-specific error rate was 1%, whereas the context-independent error was 9%. A method for merging isomorphic networks into a single network is described. The method uses singular value decomposition to find a minimal basis for the set of context-specific weight vectors. Context-dependent linear combinations of the basis vectors may then be computed using second-order network units. Compact networks can thus be obtained in which the vowel discrimination surfaces are modulated by the phonetic context. In a preliminary experiment, as the number of basis vectors was reduced from 6 to 3, the error rate increased from 1% to 3%. Experiments with nonlinear optimization of context-modulated, second-order networks for this task are underway and will be reported.