Phonetic Identification Research Articles

Phonetic identification in quiet and in noise by listeners with cochlear implants

This study examined the effect of noise on the identification of four synthetic speech continua (/ra/-/la/, /wa/-/ja/, /i/-/u/, and say-stay) by adults with cochlea implants (CIs) and adults with normal-hearing (NH) sensitivity in quiet and noise. Significant group-by-SNR interactions were found for endpoint identification accuracy for all continua except /i/-/u/. The CI listeners showed the least NH-like identification functions for the /ra/-/la/ and /wa/-/ja/ continua. In a second experiment, NH adults identified four- and eight-band cochlear implant stimulations of the four continua, to examine whether group differences in frequency selectivity could account for the group differences in the first experiment. Number of bands and SNR interacted significantly for /ra/-/la/, /wa/-/ja/, and say-stay endpoint identification; strongest effects were found for the /ra/-/la/ and say-stay continua. Results suggest that the speech features that are most vulnerable to misperception in noise by listeners with CIs are those whose acoustic cues are rapidly changing spectral patterns, like the formant transitions in the /wa/-/ja/ and /ra/-/la/ continua. However, the group differences in the first experiment cannot be wholly attributable to frequency selectivity differences, as the number of bands in the second experiment affected performance differently than suggested by group differences in the first experiment.

The preservation of schwa in the converging phonological system of Frenchville (PA) French

The phonological system of the French of Frenchville, Pennsylvania (USA) demonstrates a dramatic case of transfer in the latest (and last) generation of bilingual French–English speakers: the mid front round vowels, [œ] and [ø], have often been replaced by the English rhoticized schwa as found in the word sir. However, French schwa, which is arguably phonetically non-distinct from the mid front round vowels, does not participate fully in this merger. This result is unexpected given both the phonetic identity of schwa and [ø], and the fact that our speakers are not literate in French and, as such, have no access to the differential orthographic representations manifest between schwa and the mid front round vowels. The data argue strongly that schwa is, in some sense, “real” for these speakers. Based on a phonetic analysis of the vowels under consideration, we argue that transfer between two sound systems cannot be perceived as a simple case of phonetic replacement. Instead, transfer or convergence with English must be viewed as a systemic process that preserves contrast in unexpected ways. In the case at hand, the data suggest that the traditional separation between the phonetic and phonological levels of grammar cannot be maintained as each level contributes to both provoking merger and maintaining contrast in bilingual speech.

A phonetically neutral model of the low-level audio-visual interaction

The improvement of detectability of visible speech cues found by Grant and Seitz [2000. The use of visible speech cues for improving auditory detection of spoken sentences. JASA 108, 1197–1208] has been related to the degree of correlation between acoustic envelopes and visible movements. This suggests that audio and visual signals could interact early during the audio-visual perceptual process on the basis of audio envelope cues. On the other hand, acoustic-visual correlations were previously reported by Yehia et al. [1998. Quantitative association of vocal tract and facial behavior. Speech Commun. 26 (1), 23–43]. Taking into account these two main facts, the problem of extraction of the redundant audio-visual components is revisited: the video parametrization of natural images and three types of audio parameters are tested together, leading to new and realistic applications in video synthesis and audio-visual speech enhancement. Consistent with Grant and Seitz’s prediction, the 4-subband envelope energy features are found to be optimal for encoding the redundant components available for the enhancement task. The proposed computational model of audio-visual interaction is based on the product, in the audio pathway, between the time-aligned audio envelopes and video-predicted envelopes. This interaction scheme is shown to be phonetically neutral, so that it will not bias phonetic identification. The low-level stage which is described is compatible with a late integration process, which may be used as a potential front-end for speech recognition applications.

Perception of Mandarin Lexical Tones when F0 Information is Neutralized

In tone languages, the identity of a word depends on its tone pattern as well as its phonetic structure. The primary cue to tone identity is the fundamental frequency (F0) contour. Two experiments explore spoken word recognition how listeners perceive Mandarin monosyllables in which all or part of the F0 information has been neutralized. In Experiment 1, tone languages supposedly critical portions of the tonal pattern were neutralized with signal processing techniques, yet identification of the tonal pattern remained quite good. In Experiment 2, even more drastic removal of tonal information was tested, using stimuli whispered by Mandarin speakers, or signal processed to remove the pitch cues. Again, performance was surprisingly good, showing that listeners can use secondary cues when the primary cue is unavailable. Moreover, a comparison of tone perception of naturally whispered monosyllables and the signal processed ones suggests that Mandarin speakers promote the utility of secondary cues when they know that the primary cue will be unavailable. The flexible use of cues to tone in Mandarin is similar to the flexibility that has been found in the production and perception of cues to phonetic identity in Western languages.

Factors that allow a high level of speech understanding by patients fit with cochlear implants.

Three factors account for the high level of speech understanding in quiet enjoyed by many patients fit with cochlear implants. First, some information about speech exists in the time/amplitude envelope of speech. This information is sufficient to narrow the number of word candidates for a given signal. Second, if information from the envelope of speech is available to listeners, then only minimal information from the frequency domain is necessary for high levels of speech recognition in quiet. Third, perceiving strategies for speech are inherently flexible in terms of the mapping between signal frequencies (i.e., the locations of the formants) and phonetic identity.

The effect of variation in naturalness on phonetic perceptual identification

The relation between apparent naturalness and phonetic identification was assessed in six perceptual tests. A seven-step place-of-articulation series spanning [da] to [ga] was created with speech synthesis approximating the spectra of natural samples. The sensitivity of perceivers to this realization of a place contrast was assessed by estimating the cumulative d′ across the series in identification tests. Four variants of this series differing in apparent naturalness were produced by altering the synthesis source function while preserving the center frequency and bandwidth of the formants, and by replicating the gross spectrotemporal patterns with time-varying sinusoids. In addition to calibrating perceivers’ sensitivity to the place contrast over variation in naturalness, we conducted a naturalness tournament composed of items drawn from the five test series. A correlation of the findings of the naturalness tournament with the measures of phonetic sensitivity offers an index of the effect of variation in naturalness on phonetic perception. This study can resolve the dispute between the classic premise that intelligibility and naturalness are orthogonal attributes of speech perception, and the more recent premise entailed by episodically based accounts of perceptual categorization, that novel instances are identified by virtue of auditory similarity to prior exemplars. [Research supported by NIDCD.]

Specifying voicing differences in children’s productions of syllable-final stops: Knowledge versus skill

Among the acoustic correlates of phonetic identity considered to be universal is the length of vocalic segments preceding syllable-final stops, which is a correlate to the voicing of those stops. However, findings reported earlier [S. Nittrouer et al., J. Acoust. Soc. Am. 109, 2312(A) (2001)] showed that the commonly described length effect (i.e., shorter segments before voiceless than before voiced stops) is attenuated in adults’ samples from continuous discourse, and that listeners of all ages fail to make much use of this effect in perceptual decisions, preferring instead to base voicing judgments on dynamic spectral information. Subsequent to that study, acoustic measures (duration of the preceding vocalic segment and frequency of the first formant at voicing offset) of children’s (5 and 7 years of age) productions of words differing in the voicing of syllable-final stops showed that by 5 years of age children’s productions generally had the same acoustic structure as those of adults, but within-speaker variability on both measures was roughly twice as great in children’s as in adults’ productions. Thus, children were trying to coordinate the vocal-tract closing and glottal abduction gestures as adults do, but were not skilled enough to do so reliably. [Work supported by NIDCD.]

Age differences in the influence of metrical structure on phonetic identification

Two phonetic identification experiments were conducted with two groups of participants: a young adult group and an older adult group. In Experiment 1, subjects were required to make voiced–voiceless decisions for initial alveolar stop consonants to stimuli along two voice onset time (VOT) continua—one ranging from “di’gress” to “ti’gress” and the other from “’digress” to “’tigress” (i.e., in one continuum, the voiced endpoint was consistent with the word’s stress pattern while in the other continuum, the voiceless endpoint was consistent with the word’s stress pattern). Results revealed that both groups of participants were influenced by the stress pattern of the stimuli, but stress seemed to override VOT cues for a large number of the older individuals. To confirm that the effect was not simply due to a lexical influence, a follow-up experiment utilized two word–nonword continua (“diamond–tiamond” and “diming–timing”) to examine the magnitude of lexical effects in these subject groups. Typical lexical status effects emerged for both young and older adults which were smaller than the effects of stress pattern found in Experiment 1. The findings are discussed with respect to the role of prosodic context in language processing in aging.

Contextual Influences on Phonetic Identification in Aphasia: The Effects of Speaking Rate and Semantic Bias

Two experiments examined the influence of context on stop-consonant voicing identification in fluent and nonfluent aphasic patients and normal controls. Listeners were required to label the initial stop in a target word varying along a voice onset time (VOT) continuum as either voiced or voiceless ([b]/[p] or [d]/[t]). Target stimuli were presented in sentence contexts in which the rate of speech of the sentence context (Experiment 1) or the semantic bias of the context (Experiment 2) was manipulated. The results revealed that all subject groups were sensitive to the contextual influences, although the extent of the context effects varied somewhat across groups and across experiments. In addition, a number of patients in both the fluent and nonfluent aphasic groups could not consistently identify even endpoint stimuli, confirming phonetic categorization impairments previously shown in such individuals. Results are discussed with respect to the potential reliance by aphasic patients on higher level context to compensate for phonetic perception deficits.

Relationship between N1 evoked potential morphology and the perception of voicing.

Auditory evoked potential (AEP) correlates of the neural representation of stimuli along a /ga/-/ka/ and a /ba/-/pa/ continuum were examined to determine whether the voice-onset time (VOT)-related change in the N1 onset response from a single to double-peaked component is a reliable indicator of the perception of voiced and voiceless sounds. Behavioral identification results from ten subjects revealed a mean category boundary at a VOT of 46 ms for the /ga/-/ka/ continuum and at a VOT of 27.5 ms for the /ba/-/pa/ continuum. In the same subjects, electrophysiologic recordings revealed that a single N1 component was seen for stimuli with VOTs of 30 ms and less, and two components (N1' and N1) were seen for stimuli with VOTs of 40 ms and more for both continua. That is, the change in N1 morphology (from single to double-peaked) coincided with the change in perception from voiced to voiceless for stimuli from the /ba/-/pa/ continuum, but not for stimuli from the /ga/-/ka/ continuum. The results of this study show that N1 morphology does not reliably predict phonetic identification of stimuli varying in VOT. These findings also suggest that the previously reported appearance of a "double-peak" onset response in aggregate recordings from the auditory cortex does not indicate a cortical correlate of the perception of voicelessness.

The obligatory use of vowel duration in speaking rate normalization

Speech segment duration varies with speaking rate and phonetic identity. This study explored how listeners normalize for speaking rate when variation in segmental duration was a result of phonetic identity. Listeners identified the initial consonant in series that ranged from ‘‘beat’’ to ‘‘wheat’’ and ‘‘bead’’ to ‘‘weed.’’ Even though the endpoints of the series were spoken at the same speaking rate, the vowel duration in ‘‘bead’’–‘‘weed’’ was longer than ‘‘beat’’–‘‘wheat.’’ Luce and Charles-Luce [J. Acoust. Soc. Am. 78, 1949–1957 (1985)] showed that vowel duration is an acoustic correlate of final stop voicing. Even though the vowel duration difference was an acoustic correlate of the final consonant voicing, listeners responded to the series as if vowel duration variation were also a cue to speaking rate. Listeners gave more /b/ responses to the ‘‘bead’’–‘‘weed’’ series. These data are inconsistent with models of perception in which acoustic correlates are uniquely assigned to sources in speech production. Instead, speaking rate normalization appears to be an early, autonomous process and the perceptual utilization of duration information for multiple aspects of speech perception may be obligatory. [Work supported by NIDCD Grant No. RO1DCOO219 to SUNY at Buffalo.]

Integrality of nasalization and F1. II. Basic sensitivity and phonetic labeling measure distinct sensory and decision-rule interactions.

In vowel perception, nasalization and height (the inverse of the first formant, F1) interact. This paper asks whether the interaction results from a sensory process, decision mechanism, or both. Two experiments used vowels varying in height, degree of nasalization, and three other stimulus parameters: the frequency region of F1, the location of the nasal pole/zero complex relative to F1, and whether a consonant following the vowel was oral or nasal. A fixed-classification experiment, designed to estimate basic sensitivity between stimuli, measured accuracy for discriminating stimuli differing in F1, in nasalization, and on both dimensions. A configuration derived by a multidimensional scaling analysis revealed a perceptual interaction that was stronger for stimuli in which the nasal pole/zero complex was below rather than above the oral pole, and that was present before both nasal and oral consonants. Phonetic identification experiments, designed to measure trading relations between the two dimensions, required listeners to identify height and nasalization in vowels varying in both. Judgments of nasalization depended on F1 as well as on nasalization, whereas judgments of height depended primarily on F1, and on nasalization more when the nasal complex was below than above the oral pole. This pattern was interpreted as a decision-rule interaction that is distinct from the interaction in basic sensitivity. Final consonant nasality had little effect in the classification experiment; in the identification experiment, nasal judgments were more likely when the following consonant was nasal.

The Influence of Neighborhood Density on Phonetic Categorization in Aphasia

The present study examined the contribution of lexically based sources of information to acoustic–phonetic processing in fluent and nonfluent aphasic subjects and age-matched normals. To this end, two phonetic identification experiments were conducted which required subjects to label syllable-initial bilabial stop consonants varying along a VOT continuum as either /b/ or /p/. Factors that were controlled included the lexical status (word/nonword) and neighborhood density values corresponding to the two possible syllable interpretations in each set of stimuli. Findings indicated that all subject groups were influenced by both lexical status and neighborhood density in making phonetic categorizations. Results are discussed with respect to theories of acoustic–phonetic perception and lexical access in normal and aphasic populations.

Top-down influences on phonetic processing as a function of aging

A phonetic identification experiment was conducted in order to assess the extent to which lexical information influenced low-level speech perception processes in three different age groups of subjects (20–29 years, 60–69 years, and 70–79 years). Subjects were required to label syllable-initial, bilabial stop consonants which varied in voice onset time (VOT) as either /b/ or /p/. Following Ganong [J.E.P.: H.P.P. 6, 110–125 (1980)], a pair of VOT continua with opposing lexical biases was created: one continuum ranged from the word /buk/ to the nonword /puk/ while the other continuum ranged from the nonword /bu t/ to the word /put/. VOT values for the syllable-initial /b/–/p/ phoneme were the same at corresponding steps in the two continua. Results indicated a significant influence of lexical status on phonetic identifications for all age groups. Furthermore, a significant difference in the location of the phonetic crossover boundary was observed, with older subjects showing longer VOT values than young subjects. While findings suggest that top-down processing is still a prevalent part of phonetic perception in elderly individuals, it appears that aging may bring changes in the values of the acoustic parameters used to define phonetic categories. [Work supported by McGill University.]

Comparing models for audiovisual fusion in a noisy-vowel recognition task

Audiovisual speech recognition involves fusion of the audio and video sensors for phonetic identification. There are three basic ways to fuse data streams for taking a decision such as phoneme identification: data-to-decision, decision-to-decision, and data-to-data. This leads to four possible models for audiovisual speech recognition, that is direct identification in the first case, separate identification in the second one, and two variants of the third early integration case, namely dominant recoding or motor recoding. However, no systematic comparison of these models is available in the literature. We propose an implementation of these four models, and submit them to a benchmark test. For this aim, we use a noisy-vowel corpus tested on two recognition paradigms in which the systems are tested at noise levels higher than those used for learning. In one of these paradigms, the signal-to-noise ratio (SNR) value is provided to the recognition systems, in the other it is not. We also introduce a new criterion for evaluating performances, based on transmitted information on individual phonetic features. In light of the compared performances of the four models with the two recognition paradigms, we discuss the advantages and drawbacks of these models, leading to proposals for data representation, fusion architecture, and control of the fusion process through sensor reliability.

Depolarizing the perceptual magnet effect.

In recent years there has been a great deal of interest in demonstrations of the so-called "Perceptual-Magnet Effect" (PME). In these studies, AX-discrimination tasks purportedly reveal that discriminability of speech sounds from a single category varies with judged phonetic "goodness" of the sounds. However, one possible confound is that category membership is determined by identification of sounds in isolation, whereas, discrimination tasks include pairs of stimuli. In the first experiment of the current study, identification and goodness judgments were obtained for vowels (/i/-/e/) presented in pairs. A substantial shift in phonetic identity was evidenced with changes in the context vowel. In a second experiment, listeners participated in an AX-discrimination task with the vowel pairs from the first experiment. Using the contextual identification functions from the first experiment, predictions of discriminability were calculated using the classic tenets of Categorical Perception. Obtained discriminability functions were well accounted for by predictions from identification. There was no additional unexplained variance that required the proposal of "perceptual magnets." These results suggest that PME may be nothing more than further demonstration that general discriminability is greater for cross-category stimulus pairs than for within-category pairs.

Perception of American English /r/ and /l/ by Mandarin speakers: Influences of phonetic identification and category goodness

Studies in speech perception show a perceptual magnet effect for listeners which is characterized by shrunken perceptual distances near excellent exemplars of phonetic categories and stretched distances near poor exemplars. This study extended this work by examining the influence of Mandarin speakers’ phonetic identification and category goodness on their perception of American English /r/ and /l/. Eighteen /ra/ and /la/ tokens were used that varied F2 and F3 frequencies. Listeners identified stimuli as Mandarin /u/, /m/, or /l/, provided goodness ratings for them, and rated the similarity of all stimulus pairs. Multidimensional scaling analyses of similarity ratings exhibited that the perceptual ‘‘map’’ for Mandarin speakers differs substantially from that of American speakers [Iverson and Kuhl, J. Acoust. Soc. Am. (1996)] listening to the same stimuli. The results indicate that (1) Mandarin speakers perceived /u/ and /l/ in the space, (2) perceptual distance was shrunk near the best exemplars of phonetic categories and stretched near the category boundary, (3) variations of F2 were more important than F3, and (4) individual differences in /l/ identification corresponded to a degree of shrinking near the best exemplars of the /l/ category. Results indicate that listeners use their native-language representations in the perception of a non-native language.

On the Strategic Weighting of Spectral Information in Consonant Identification

When identifying speech sounds, do listeners direct attention specifically to those frequency regions in which distinctive phonetic information is located? Scharf, Dai, and Miller (1988) concluded that they do, based on an experiment that involved listeners identifying the members of a pair of voiced stop consonants differing primarily in their third-formant transitions. Adding extra contrastive acoustic energy to the third-formant region of one of the consonants enhanced performance more than did adding energy to a region where the consonants were otherwise identical. The present experiments extended this paradigm to reveal evidence of an interaction between the frequency region in which extra distinctive energy was added and the class of consonant to which it was added. The results were more consistent with an account in terms of the effects of added energy on phonetic identity than one based on spectral selective attention.

Sensorimotor adaptation in speech production.

Human subjects are known to adapt their motor behavior to a shift of the visual field brought about by wearing prism glasses over their eyes. The analog of this phenomenon was studied in the speech domain. By use of a device that can feed back transformed speech signals in real time, subjects were exposed to phonetically sensible, online perturbations of their own speech patterns. It was found that speakers learn to adjust their production of a vowel to compensate for feedback alterations that change the vowel's perceived phonetic identity; moreover, the effect generalizes across phonetic contexts and to different vowels.

Phonological impairment in dyslexic children with and without early speech-language disorder.

This study examines phonological awareness in a group of 10 dyslexic children, compared with two groups of children (a reading-equivalent control group and a group of beginning readers). Five of the dyslexic children exhibited an early speech-language impairement, and five others were not language-impaired. The experimental design consisted of a set of 10 tasks involving sensitivity to phonological strings, phonetic identification, and phoneme segmentation and manipulation. The major findings reveal that (a) the two subgroups of dyslexic children exhibit a metalinguistic impairment and (b) dyslexic children with speech-language impairment exhibit (besides metalinguistic impairment) a deficit concerning early precursors of the phonological awareness (rhyme and syllables). These findings provide additional support for including speech-language impairment as an operational criterion within the heterogeneous group of dyslexic children.