Syllabic Context Research Articles

Hearing acoustic details of speech. III. Discrimination of fricative duration

Listeners' ability to dicriminate the duration of the frication noise in the syllable /t∫a/ was measured. An 842‐ms token of /t∫a/ produced by a female talker was digitized at 20 kHz, output via a D/A converter and presented on each interval of a 2IFC task. On each trial, one of the intervals contained a 5–35‐ms decrement in the duration of the 183‐ms frication noise. Duration decrements were obtained by removing segments from the temporal center of the noise. Normal‐hearing listeners decided whether the first or the second interval contained the shorter‐duration frication noise. Duration discrimination was studied for frication noise presented in isolation and in syllabic context. Duration‐discrimination thresholds [P(c) = 75] were interpolated from probits fitted to 7‐point psychometric functions. On the average, thresholds in isolation were ∼14 ms or ∼8% of the duration of the frication noise. Thresholds in context were ∼32.0 ms or ∼18% of the fricative duration. Although larger than in isolation, thresholds in context are smaller than 8% of the overall syllable duration. [Work supported by OCAST Grant HSO‐005 and Presbyterian Health Foundation.]

Syllable structure and units of analysis in speech perception

Syllables or their constituents, such as onsets and rhymes, may be basic units in speech perception. If so, then recognition of a target phoneme might be influenced more by its syllabic context (i.e., other phonemes in the same syllable or constituent) than by other aspects of phonetic context. To investigate the influence of syllable structure on phoneme perception, subjects in three response-time experiments were instructed to identify a target phoneme, ignoring variation in an adjacent phoneme. In one condition, the adjacent phoneme was held constant; in a second condition, contextual variation was irrelevant to target identity. In two experiments, context varied either within the same syllabic onset or in an adjacent syllable. The third experiment involved varying an adjacent phoneme either in the same syllable as the target or in a neighboring syllable. If the syllable or onset is perceived as a unit, then varying context within that unit should slow target recognition more than varying context outside the syllable. The results have implications for the role of syllable structure in speech perception. [Research supported by AFOSR and BRSG.]

Phonological parsing and lexical retrieval

Speech is a natural error-correcting code. The speech signal is full of rich sources of contextual redundancy at many levels of representation including allophonic variation, phonotactics, syllable structure, stress domains, morphology, syntax, semantics and pragmatics. The psycholinguistic literature has tended to concentrate heavily on high level constraints such as semantics and pragmatics and has generally overlooked the usefulness of lower level constraints such as allophonic variation. It has even been said that allophonic variation is a source of confusion or a kind of statistical noise that makes speech recognition that much harder than it already is. In contrast, I argue that aspiration, stop release, flapping, palatalization and other cues that vary systematically with syllabic context can be used to parse syllables and stress domains. These constituents can then constrain the lexical matching process, so that much less search will be required in order to retrieve the correct lexical entry. In this way, syllable structure and stress domains will be proposed as an intermediate level of representation between the phonetic description and the lexicon. My argument is primarily a computational one and will include a discussion of a prototype phonetic parser which has been implemented using simple well- understood parsing mechanisms. No experimental results will be presented.

Vowel recognition in the absence of formant cues: Dynamic contributions to perception

The primary acoustic cue to vowel identity has typically been related to the center frequencies of the first two or three vowel formants. In experiments using specially degraded stimuli, results consistent with this assumption were obtained for stationary synthetic vowels, but not for vowels presented in syllabic contexts (e.g., /VwV/ and /ərVd/). For these experiments, vowel stimuli (both stationary and in context) were synthesized with high‐frequency square waves substituted for the standard synthesizer voicing source (Klatt, 1980 cascade synthesizer). Spectra associated with these stimuli have peaks, due to the source harmonics, located at odd multiples of the fundamental frequency. In identification studies using the vowels [ /i/, /ɪ/, /ε/, /æ/, /a/ ], listeners tended to identify stationary vowels on the basis of the frequencies of their largest‐magnitude spectral peaks. Since such peaks were invariably coincident with source harmonics, identification errors were frequent. Errors were, however, restricted almost entirely to the vowels [/ɪ/, /ε/, /æ/]. By contrast, vowels in syllabic context were reliably more resistent to identification errors, despite nearly identical acoustic structure throughout the medial portion of the vowel. Results are interpreted to suggest that listeners are well attuned to the contrast between source and vocal tract contributions to the signal. In order to separate these two contributions, however, listeners appeared to need information about changes over time in at least one function (the vocal tract function for these stimuli). [Work supported by NINCDS and the University of Maryland.]

Influence of following context on perception of the voiced—voiceless distinction in syllable‐final stop consonants

The duration of the vowel preceding a syllable‐final stop consonant (as defined by some acoustic criterion) generally is a potent cue to the perception of the stop as phonologically voiced or voiceless. We show that the absolute vowel duration that characterizes the voiced—voiceless boundary in a monosyllable is reduced considerably when a second syllable beginning with a different stop consonant is added to form a disyllabic word. We also show that this contextual effect declines with increasing temporal separation between the two syllables, and we examine this decline in relation to listener's judgments of whether they hear the stimulus as one disyllabic word or as two successive monosyllabic words. The effect of syllabic context, apart from introducing closure duration as an additional voicing cue, may reflect listeners' tacit knowledge of the temporal compression that occurs in speech production as the number of syllables in a word is increased. [Research supported by NICHD.]

Effects of consonant context on the perception of French vowels

American English vowels have been found to be more accurately identified in consonantal context than in isolation. One might, however, expect different results for French because phonological and phonetic characteristics of French vowels are different from those of English. Native French listeners identified stressed oral vowels produced by native speakers in /t/-vowel and isolated vowel contexts more accurately than vowels produced in vowel-/t/ and /t/-vowel-/t/ contexts. In categorial discrimination, native French, non-native speakers of French, and non-speakers of French perceived isolated French vowels more accurately than vowels in TVT context, arguing against phonological appropriateness as the determinant of context effect differences between languages. American non-native speakers of French and non-speakers of French were significantly less accurate than native French in identification and discrimination of French vowels in all contexts, especially front rounded vowels which do not occur in English. Acoustical analyses of the French vowels showed less diphthongization than for English vowels in the same syllabic contexts, making vowel “targets” more reliable as sources of information for French vowel identity.

The role of "chirp" identification in duplex perception.

Duplex perception occurs when the phonetically distinguishing transitions of a syllable are presented to one ear and the rest of the syllable (the “base”) is simultaneously presented to the other ear. Subjects report hearing both a nonspeech “chirp” and a speech syllable correctly cued by the transitions. In two experiments, we compared phonetic identification of intact syllables, duplex percepts, isolated transitions, and bases. In both experiments, subjects were able to identify the phonetic information encoded into isolated transitions in the absence of an appropriate syllabic context. Also, there was no significant difference in phonetic identification of isolated transitions and duplex percepts. Finally, in the second experiment, the category boundaries from identification of isolated transitions and duplex percepts were not significantly different from each other. However, both boundaries were statistically different from the category boundary for intact syllables. Taken together, these results suggest that listeners do not need to perceptually integrate F2 transitions or F2 and F3 transition pairs with the base in duplex perception. Rather, it appears that listeners identify the chirps as speech without reference to the base.

Perception of French vowels by native French and American English listeners

Studies of American English vowel identification have found that vowels are better identified in consonantal context than in isolation. Strange, Edman, and Jenkins [J. Exp. Psych.: Hum. Percept. Perform. 5, 643–656 (1979)] attribute this finding to dynamic acoustic information carried by formant transitions and vowel duration. One might expect different results for French, because phonological constraints and phonetic characteristics of French vowels are different from those of English. Four native Parisian French speakers produced 11 stressed, oral French vowels in four syllabic contexts: /t/‐vowel‐/t/ (TVT), vowel‐/t/ (VT), /t/‐vowel (TV), and isolated vowel (♯V♯). Native French speakers identified vowels in TV and ♯V♯ contexts more accurately than vowels in VT and TVT contexts. In a categorial ABX discrimination task, non‐native French speakers and nonspeakers of French also perceived isolated French vowels better than vowels in TVT context, arguing against phonological interpretations of the differences between languages. Acoustic factors accounting for differences in French and English vowel perception are discussed. [Supported by NIMH, NICHHD, NSF, and Graduate School of University of Minnesota.]

Acoustic and phonological factors in vowel identification.

Vowels are better identified in a consonantal syllabic context than as isolated vowels. This finding is contrary to predictions from traditional theories of vowel perception. The poor perception of isolated vowels might be attributed to a lack of dynamic acoustic cues or to familiarity effects related to the phonological rules of English. Vowel identification tests were conducted using six talkers, nine vowels, and seven syllabic contexts. Consonantal context improved vowel identification; final consonants aided identification more than initial consonants. No consistent support was found for the effect of phonological rules but duration information was seen to play a critical role. Results constitute a challenge to traditional theories of vowel perception.

Acoustic invariance for place of articulation in stops and nasals across syllabic contexts

In earlier ASA meetings, we presented theoretical and experimental support for the view that invariant cues for place of articulation in initial voiced stop consonants reside in the gross shape of the short-term spectrum at consonantal release. It is postulated that this concept of invariant spectral shapes is generalizable across phonetic contexts to all stop consonants and nasals. We have examined the spectral shapes for a large number of tokens of initial and final voiced and voiceless stop consonants and nasals spoken in various vowel environments by four male and two female speakers. The spectra obtained were matched to a set of templates or “windows” we developed to test quantitatively for compactness and diffuseness. Between 75% and 80% of the stop consonants were correctly classified by these templates. The failure of the remaining spectra to fit these templates may reflect limitations of the particular spectral analyses as well as the role of secondary context-dependent cues for place of articulation. The application of these procedures to nasals indicates that onsets may need to be defined in terms of changes in spectral amplitude from the nasal murmur to consonantal release. [Work supported in part by NINCDS and the John Simon Guggenheim Foundation.]

Vowel perception in consonantal context

Consonant‐vowel‐consonant syllables manifest formant frequencies in the center of the vowel region that differ from those of the vowels spoken in isolation [K.N. Stevens and A.S. House, J. Speech Hear. Res. 6, 111–128 (1963)]. It has been reported that subjects adjust their categorization of the [U]‐[I] vowel continuum in consonantal contexts in a way that compensates for the undershoot of vowel target frequencies [B.E.F. Lindblom and M. Studdert‐Kennedy, J. Acoust. Soc. Am. 42, 830–843 (1967)]. Do subjects hear the context‐embedded vowel as having formant frequencies that are not in fact reached or do they merely adjust their categorization criteria? Listeners were asked to adjust the first two formant frequencies of a synthesized steady vowel to match the color of the vowel heard in the syllabic context. The results do not in general support the undershoot hypothesis. Some significant shifts from the nuclear‐vowel formant frequency values to the perceived formant‐frequency values are noted in the opposite direction. The result suggest that instead of indicating extrapolated target values, the matched formant values correspond to some appropriate time‐average of the time‐varying formant frequencies. [Work supported by ARPA, D. of Defense, and NIH.]

Perception of dichotically presented vowels.

Listeners heard vowels in syllabic contexts, presented dichotically at S/N (signal-to-noise) ratios of 0 and −10 dB, and their responses were analyzed in terms of individual ears. In the most-adverse listening condition the level of correct response in the right ear was 6% higher, on the average, than that obtained in the left ear; there was no difference between ears in the more-favorable listening condition.

An Analysis of Perceptual Confusions in Identification of Dichotically Presented CVC Syllables

Earlier studies with synthetic speech have shown that the method of presenting pairs of competing syllables dichotically yields reliable differences in the accuracy of recognition of material presented to the left and right ears. Contrasting consonants, presented in consonant-vowel (CV) syllables as minimal pairs, are more accurately identified from the right ear than from the left. A significant right-ear advantage was not found when steady-state vowels were presented in isolation. Since there are a number of possible explanations for this difference in perception of vowels and consonants, further experiments were performed using “real speech” with vowels in syllabic context. The results were essentially the same as in the earlier experiment—consonants were significantly better identified from the right-ear channel; vowels gave no consistent lateral differences. These results point to a genuine difference in the mode of central processing of consonants and vowels. An analysis of the matrix of confusions was carried out in order to discover what feature information is retained and what is lost in listening to syllables presented dichotically.

Confusions of Gravity and Diffuseness of Filtered Prevocalic /p,t,k/ and Postconsonantal /u,i,a/

The relationship of /p/-/t/ according to Jakobson carries the opposition grave (low pitch) and acute (high pitch). Fischer-Jørgensen found that this opposition corresponds to a physical difference of relatively low and high frequencies. Further, compactness differentiates /p/ and /t/ from /k/. This opposition is relative to the concentration of energy in a narrow central part of either the higher or lower portion of the spectrum. The vowels /u,i,a/ are similarly contrasted. This experiment tests the relativeness of grave and acute and the distributions of spectrum for diffuseness. The speech spectrums of /p,t,k/ in a prevocalic syllabic context /u,i,a/ were recorded by one male and one female and filtered in 102 different frequency bands. Native speakers (English, N=205) listened in groups of five. Each group heard two bands in an open-choice situation at different times. The data were analyzed separately for correct identification and for substitutions. Contextual effects on perception were demonstrated. Optimal bands were also determined. Substitutions indicated general agreement but failed to demonstrate clearly the Jakobsonian distribution of relative frequency ranges. Optimal-intelligibility bands were found over a wide range of the spectrum.