Stop Consonant Place Research Articles

Classification of stop place in consonant-vowel contexts using feature extrapolation of acoustic-phonetic features in telephone speech

Knowledge-based speech recognition systems extract acoustic cues from the signal to identify speech characteristics. For channel-deteriorated telephone speech, acoustic cues, especially those for stop consonant place, are expected to be degraded or absent. To investigate the use of knowledge-based methods in degraded environments, feature extrapolation of acoustic-phonetic features based on Gaussian mixture models is examined. This process is applied to a stop place detection module that uses burst release and vowel onset cues for consonant-vowel tokens of English. Results show that classification performance is enhanced in telephone channel-degraded speech, with extrapolated acoustic-phonetic features reaching or exceeding performance using estimated Mel-frequency cepstral coefficients (MFCCs). Results also show acoustic-phonetic features may be combined with MFCCs for best performance, suggesting these features provide information complementary to MFCCs.

Grounding stop place systems in the perceptuo-motor substance of speech: On the universality of the labial–coronal–velar stop series

Vowels are by far the best understood units in human sound systems, and are well characterized at the articulatory, acoustic, and perceptual levels. This has permitted explanations of vowel systems as structured by perception, and has led to effective substance-based theories. By contrast, stops are far less thoroughly understood. In this paper we use an articulatory–acoustic model of the vocal tract to examine stop consonant place in terms of both articulation and formant values. This allows us to locate each place of articulation in the F1–F2–F3 space, and to demonstrate in “articulatory nomograms” how formants evolve while closure is displaced from the front to the back of the vocal tract. Then, in the framework of the “Perception for Action Control Theory” that we have developed in recent years, we show that the near universal labial–coronal–velar stop series (i.e., /b d ɡ/ or /p t k/) is a perceptually optimal structure for stops just as /i a u/ is for vowels, provided that it is embedded in a suitable perceptuo-motor framework.

Acoustic-phonetic features for stop consonant place detection in clean and telephone speech

This work classifies voiceless stop consonant place in CV tokens of English using burst release cues for clean (TIMIT) and telephone speech (NTIMIT). We compared the performance of cepstral coefficients to acoustic phonetics-motivated features such as center of gravity, burst amplitude and relative difference of formant amplitudes. In clean speech, cepstral coefficients resulted in better classification. However, for test data from NTIMIT, acoustic phonetic-based features outperformed cepstral coefficients, particularly if models were trained on clean speech. In addition, augmenting cepstral coefficients with acoustic phonetic-based measurements resulted in the best performance. These findings suggest that cepstral coefficients are able to model speech in a given environment in finer detail, whereas acoustic phonetic-based features are more robust to changes in environment, so that combining both types of measurements leads to the best performance.

Multichannel Compression Hearing Aids: Perceptual Considerations

Multichannel Compression Hearing Aids: Perceptual Considerations

Perceptual Weighting of Stop Consonant Cues by Normal and Impaired Listeners in Reverberation Versus Noise

To determine if listeners with normal hearing and listeners with sensorineural hearing loss give different perceptual weightings to cues for stop consonant place of articulation in noise versus reverberation listening conditions. Nine listeners with normal hearing (23-28 years of age) and 10 listeners with sensorineural hearing loss (31-79 years of age, median 66 years) participated. The listeners were asked to label the consonantal portion of synthetic CV stimuli as either /p/ or /t/. Two cues were varied: (a) the amplitude of the spectral peak in the F4/F5 frequency region of the burst was varied across a 30-dB range relative to the adjacent vowel peak amplitude in the same frequency region, (b) F2/F3 formant transition onset frequencies were either appropriate for /p/, /t/ or neutral for the labial/alveolar contrast. Weightings of relative amplitude and transition cues for voiceless stop consonants depended on the listening condition (quiet, noise, or reverberation), hearing loss, and age of listener. The effects of age with hearing loss reduced the perceptual integration of cues, particularly in reverberation. The effects of hearing loss reduced the effectiveness of both cues, notably relative amplitude in reverberation. Reverberation and noise conditions have different perceptual effects. Hearing loss and age may have different, separable effects.

F2 Locus Equations: Phonetic Descriptors of Coarticulation in 17- to 22-Month-Old Children

The general purpose of this research was to describe coarticulation across voiced stop consonant place of articulation in 10 children younger than 2 years of age. A total of 1,182 voiced stop CV productions was analyzed using the locus equation metric, which yielded 3 regression lines that described the relation of F2 onset and F2 vowel for /bV/, /dV/, and /gV/ productions. The results revealed significant differential effects for slope and y-intercept as a function of stop consonant place of articulation. The ordering of the mean slope values for stop consonant place of articulation was /g/>/b/ and /d/, indicating that /g/ was produced with significantly greater coarticulation than /b/ or /d/. However, the unique vowel allophonic pattern of [g] coarticulation reported in the literature for English-speaking adults was generally not learned by these young children. Group and individual coarticulation trends are described in relation to developmental theories of sound acquisition. Results suggest that early coarticulation patterns are phoneme specific.

Effects of hearing impairment on the perception and neural representation of time-varying spectral cues

Differences in phonetic boundaries versus normal controls suggest that listeners with hearing impairment (HI) have difficulty categorizing stop consonant place of articulation based solely on the dynamic spectral information present in the second formant transition (F2), even when the stimuli are amplified. This may be due to a degraded ability of the central auditory nervous system to process time-varying spectral cues despite ensuring overall audibility. However, increasing the overall level of the stimuli may not result in improved audibility of F2. To determine if spectral shaping of F2 improves performance of listeners with HI, psychometric functions and N1-P2 cortical responses were compared in 10 older listeners with normal hearing versus 10 older listeners with HI. Stimuli were synthetic consonant-vowels along a /ba/-/da/-/ga/ place-of-articulation continuum in an unshaped and shaped condition. Generally, behavioral and N1-P2 results indicate that, with shaping, categorization of /d/ and /g/ improves. These findings suggest that enhanced audibility of F2 through spectral shaping does improve perception of stop consonant stimuli. However, categorical boundaries for the individuals with HI are shifted lower in frequency with shaping for all phonemes versus normal controls, indicating that enhancing audibility improves but does not completely normalize categorization performance.

Classification of stop consonant place of articulation

In this study we develop an experimental procedure for examining the relative importance of knowledge-based cues for identifying place of articulation for stop consonants. A set of acoustic attributes is selected for place classification of stops: amplitude and energy of burst, formant movement of adjacent vowels, spectrum of noise after the release, and some temporal cues. The ability of each attribute to separate the three places is evaluated by the classification error based on the distributions of its values for the three places, and another quantifier based on F ratio. These two quantifiers generally agree and show how well each individual attribute separates the three places. Linear discriminant function analysis is used to address the relative importance of these attributes when combinations are used. Their discriminating abilities and the ranking of their relative importance to the classification in different vowel and voicing contexts are reported. The overall findings are that attributes relating to the burst spectrum in relation to the vowel contribute most effectively, while formant transition is somewhat less effective. The approach used in this study can be applied to different classes of sound, as well as stops in different noise environments. [Work supported by NIH Grant Number DC 02978.]

Perceptual weighting of relative amplitude and formant transition cues in aided CV syllables.

The current study explored the changes in weighting of relative amplitude and formant transition cues that may be caused by a K-amp circuit. Twelve listeners with normal hearing and 3 listeners with sensorineural hearing loss labeled the stop consonant place of articulation of synthetic consonant-vowel stimuli. Within the stimuli, two acoustic cues were varied: the frequency of the onset of the second and third formant (F2/F3) transitions and the relative amplitude between the consonant burst and the following vowel in the fourth and fifth formant (F4/ F5) frequency region. The variation in the two cues ranged from values appropriate for a voiceless labial stop consonant to a voiceless alveolar stop consonant. The listeners labeled both the unaided stimuli and the stimuli recorded through a hearing aid with a K-amp circuit. An analysis of variance (ANOVA) model was used to calculate the perceptual weight given each cue. Data from listeners with normal hearing show a change in relative weighting of cues between aided and unaided stimuli. Pilot data from the listeners with hearing loss show a more varied pattern, with more weight placed on relative amplitude. These results suggest that calculation of perceptual weights using an ANOVA model may be worthwhile in Future studies examining the relationship between acoustic information presented by a hearing aid and the subsequent perception by the listener with hearing loss.

Effect of a Single-Channel Wide Dynamic Range Compression Circuit on Perception of Stop Consonant Place of Articulation

Previous studies have shown that altering the amplitude of a consonant in a specific frequency region relative to an adjacent vowel's amplitude in the same frequency region will affect listeners' perception of the consonant place of articulation. Hearing aids with single-channel, fast-acting wide dynamic range compression (WDRC) alter the overall consonant-vowel (CV) intensity ratio by increasing consonant energy. Perhaps one reason WDRC has had limited success in improving speech recognition performance is that the natural amplitude balances between consonant and vowel are altered in crucial frequency regions, thus disturbing the aforementioned amplitude cue for determining place of articulation. The current study investigated the effect of a WDRC circuit on listeners' perception of place of articulation when the relative amplitude of consonant and vowel was manipulated. The stimuli were a continuum of synthetic CV syllables stripped of all place cues except relative consonant amplitudes. Acoustic analysis of the CVs before and after hearing aid processing showed a predictable increase in high-frequency energy, particularly for the burst of the consonant. Alveolar bursts had more high-frequency energy than labial bursts. Twenty-five listeners with normal hearing and 5 listeners with sensorineural hearing loss labeled the consonant sound of the CV syllables in unaided form and after the syllables were recorded through a hearing aid with single-channel WDRC. There were significantly more listeners who were unable to produce a category boundary when labeling the aided stimuli. Of those listeners who did yield a category boundary for both aided and unaided stimuli, there were significantly more alveolar responses for the aided condition. These results can be explained by the acoustic analyses of the aided stimuli.

Improving consonant intelligibility for Ineraid patients fit with continuous interleaved sampling (CIS) processors by enhancing contrast among channel outputs.

In Experiment 1 the objective was to determine whether patients who have been implanted with the Ineraid electrode array perform better on tests of consonant identification when signals are processed through a continuous interleaved sampling (CIS) processor than when signals are processed through an analogue (Ineraid) processor. In Experiment 2 the objective was to determine, for patients using the CIS strategy, whether identification accuracy for stop consonant place of articulation could be improved by enhancing differences in the patterns of the signal processor channel outputs. In Experiment 1, 16 consonants were presented in VCV format for identification. In Experiment 1 the CIS patients evidence difficulty in identifying /p t k/. Therefore, in Experiment 2 the voiceless stop consonants were presented in two stimulus conditions. In one, the stimuli were unfiltered. In the other, the stimuli were individually filtered so as to enhance the differences in channel outputs for /p/, /t/, and /k/. In Experiment 1 the patients performed better with CIS processors than with analogue processors. In Experiment 2 the "enhanced" stimuli were identified with better accuracy than were the unfiltered stimuli. We confirm that Ineraid patients achieve higher scores on tests of consonant identification when using a CIS processor than when using an analogue processor. Errors in identification of stop consonant place of articulation, when using a CIS processor, are due to the similarity in the patterns of the processor's channel outputs. By showing that consonant intelligibility can be improved by filtering, we show that we have not reached the limit of speech understanding that can be supported by the population of neural elements remaining in our patients' auditory systems.

Tonotopic organization of responses reflecting stop consonant place of articulation in primary auditory cortex (A1) of the monkey

Current source density and multiunit activity elicited by stop consonant-vowel syllables were examined in primary auditory cortex of an awake monkey. Relative amplitudes of the speech-evoked responses were determined by the onset spectra of the consonants and the tonotopic organization. This finding supports the psychoacoustic hypothesis that the onset spectra of stop consonants are important determinants for the discrimination of place of articulation.

Self-organizing maps of stop consonant place from token-level locus equation inputs

Previously [D. E. Fruchter, J. Acoust. Soc. Am. 95, 2977 (1994)], identification curves were estimated for English /b,d,g/ using synthetic CV stimuli comprehensively sampling the F2-onset XF2-vowel acoustic space in the vicinity of Sussman’s /b,d,g/ locus equations. These results were used to delineate ‘‘identification surfaces’’ situated in locus equation space. The current research uses a biologically plausible neural network (the Kohonen algorithm) to model the above perception results. This algorithm is an abstraction of the local, unsupervised map-organizing process thought to occur in the brain. The Kohonen map forms a two-dimensional representation of stop consonant place categories from F2-onset and F2-vowel inputs. This emergent representation corresponds well with the experimentally observed identification surfaces and can be used to classify novel inputs and predict phoneme boundaries and confusability regions.

Inferring degree of coarticulation and relational invariance of stop place in CV’s using locus equations

Locus equations relating target and onset frequency of F2 (F2t and F2o, respectively) have been successful in distinguishing stop consonant place of articulation independent of vowel context [e.g., Sussman etal., J. Acoust. Soc. Am. 90, 1309–1325 (1991)]. These linear equations are of the form F2o=a(F2t)+c. The slope parameter a is generally interpreted as indicating not only place of articulation, but extent of coarticulation [e.g., D. Krull, PERILUS V, 43–61 (1987)] as well. The intercept parameter c often is not interpreted explicitly, except in cases where the slope is found to be near zero. The locus equation technique is considered from several different viewpoints. Locus equations are analyzed empirically by examining changes that occur in estimated slope and intercept when a trained phonetician simulates different degrees of C-V coarticulation. Differences in extent of coarticulation are also measured using electropalatography, and these measurements are compared with slope and intercept parameters derived from the acoustic data. Finally, the Locus equation technique is compared with other methods for inferring coarticulation, including absolute comparisons of individual F2o and F2t frequency differences. [Work supported by NIH.]

A developmental study of vowel perception from brief synthetic consonant–vowel syllables

The purpose of this study was to assess the perceptual role of brief synthetic consonant–vowel syllables as cues for vowels in children and adults. Stop–consonant syllables were synthesized in the context of three vowels, [i, u open aye], with durations of 10, 30, or 46 ms. These syllables were produced with bursts, and contained formant motions and/or formant onset frequencies appropriate for each stop–consonant place of articulation. Eight children at each of five age levels, 5, 6, 7, 9, and 11, and a control group of eight adults were trained to identify the vowel as either /i/, /u/, or /open aye/. The findings showed that children and adults extracted vowel information at a generally high level from stimuli as brief as one glottal pulse (together with initial burst). However, significant effects were found for age, vowel, duration, and transition type. The age effects manifested that even quantal vowels were not identified in an adult-like manner by young children. The duration and transition-type effects revealed that vowel perception was sometimes better for longer syllables with moving formant transitions. Overall, these perceptual findings indicate that syllable identification is more salient in adults than children. [Work supported by NIH, DC00464.]

Discrimination of multichannel-compressed stop consonants

The effect of full-range multichannel compression (MCC) on stop consonant discrimination was studied. For normal-hearing subjects MCC was fit to three hypothetical flat losses, with thresholds ranging from 70 to 90 dB SPL, and one hypothetical sloped loss, with thresholds normal at 500 Hz and 90 dB SPL at 4 kHz. Hearing-impaired subjects were tested with the flat MCC as well as MCC fit to the subject’s hearing loss. The MCC had compression ratios from 1.75 to 7.00, numbers of channels from 2 to 31, and Robinson–Huntington compression [C. E. Robinson and D. A. Huntington, J. Acoust. Soc. Am. 54, 314 (1973)] with 10-ms time windows. Unprocessed stimuli and frequency-equalized linear amplification were control conditions. Deleterious effects of MCC on discrimination of voice onset time (VOT) were minimal. Normal-hearing subjects had difficulty with stop-consonant place discrimination in the most severe compression conditions. Hearing-impaired subjects needed either fitted MCC or frequency-equalized linear amplification to perform consistent place discriminations on these synthetic vowel–consonant–vowel (VCV) syllables. Complete results and their implications for the application of MCC in hearing aids will be discussed. [Work supported by Department of Veterans Affairs.]

Changing the deployment of attention to phonetic structure

Previous research on perceptual learning of synthetic speech and non-native phonetic contrasts suggests that listeners shift the focus of their attention from one set of acoustic cues to another, although this hypothesis has never been tested directly. The present study was carried out to determine whether training could alter perception of stop-consonant place of articulation of shifting attention from a transition cue to the spectral shape of the release burst. CV syllables were synthesized with transitions specifying one place of articulation and release burst shape specifying another. One group of subjects was given feedback consistent with the category specified by the transitions; the other group was given feedback consistent with the bursts. Although subjects initially tended to respond based on transition cues, feedback was effective for both groups. Furthermore, for subjects who showed learning, attentional changes also generalized to the same phonetic categories in new vowel contexts. The implications of these results for theories of speech perception and perceptual learning will be discussed.

A developmental study of the perception of onset spectra for stop consonants in different vowel environments

The purpose of this study was to assess the perceptual role of integration and dynamic time-varying spectral cues for stop consonants in children and adults. Stop-consonant syllables were synthesized in the context of two vowels, /i, ɑ /. These syllables were synthesized with and without bursts, and contained formant motions and/or formant onset frequencies appropriate for each stop-consonant place of articulation. In addition, syllable duration varied between 10 to 46 ms. Children at five age levels, 4, 5, 7, 9, and 11, and a control group of adults were trained to identify the consonant as either /b/, /d/, or /g/. Preliminary findings indicate that dynamic time-varying spectral changes are not essential for accurate identification of stop consonants across all contexts. The results also show that the integration of noise bursts and formant onset frequencies, and syllable duration improve identification of alveolar and velar stops, particularly for children. The findings will be discussed relative to: (1) the differences between children and adults in processing integration and dynamic time-varying spectral cues, and (2) the developmental role of stimulus duration in the perception of stop consonants. [Work supported by NIH, DC00464.]

Perceptual identification of ambiguous consonant onsets

Theories of acoustic invariance propose that stop consonant place identification can be made independently of full consonant‐transition information. Consonant and vowel identification is possible within a few milliseconds of stop release, although the distribution of the acoustic energy in the consonant onset spectrum is frequently inconsistent with the distribution for the following vowel. The presence of oral physiologic impairment may be expected to influence whatever consonant‐vowel interaction may contribute to syllable perception. The purpose of this study was to examine consonant/vowel identification for two glossectomee speakers and one normal control. For the most impaired speaker, removal of poorly identified vowel information enhanced consonant identification. For the normal speaker and the less impaired speaker, full CVs were well identified. Removal of vowel information diminished correct consonant identification for the glossectomee speaker but for the normal speaker affected only those consonants that had spectra consistent with the following vowel. The data show that when onset spectra are ambiguous, the vowel spectrum weights the onset spectrum to influence identification of the intended consonant place of articulation.

Speech perception and frequency discrimination in good and poor readers

AbstractThe relation between speech perception and reading ability was investigated. Identification and discrimination tests with stimuli on a stop consonant place of articulation continuum (/PA/-/IA/) were presented to various groups of good and poor readers (normal first graders on a reading continuum in Study 1, moderately and severely dyslexic and high- and low-achieving normal children in Study 2). In both studies, reading-related perception differences were especially marked in a comparison of actual and predicted (from identification test) discrimination scores, suggesting that the poor readers have special difficulty with a discrimination-specific task demand. In Study 3, reading-group differences still show up if the stimulus pairs of the discrimination test are simply to be repeated (eliminating the similarity judgment factors), and are also very pronounced on a nonspeech discrimination test (with 130 ms pure tones of slightly different frequencies). In Studies 1 and 3 identification slope and phoneme boundary differences between reader groups were found as well. Though these results do not definitely prove the auditory perception hypothesis, they support it. The conditions that should be satisfied by auditory perception tasks in future reading acquisition research are briefly discussed.