Search

When a talker believes that the listener is likely to have speech perception difficulties due to a hearing loss, background noise, or a different native language, she or he will typically adopt a clear speaking style. Previous research has established that, with a simple set of instructions to the talker, ‘‘clear speech’’ can be produced by most talkers under laboratory recording conditions. Furthermore, there is reliable evidence that adult listeners with either impaired or normal hearing typically find clear speech more intelligible than conversational speech. Since clear speech production involves listener‐oriented articulatory adjustments, a careful examination of the acoustic‐phonetic and perceptual consequences of the conversational‐to‐clear speech transformation can serve as an effective window into talker‐ and listener‐related forces in speech communication. Furthermore, clear speech research has considerable potential for the development of speech enhancement techniques. After reviewing previous and current work on the acoustic properties of clear versus conversational speech, this talk will present recent data from a cross‐linguistic study of vowel production in clear speech and a cross‐population study of clear speech perception. Findings from these studies contribute to an evolving view of clear speech production and perception as reflecting both universal, auditory and language‐specific, phonological contrast enhancement features.

How does native language experience with a tone or nontone language influence pitch perception? To address this question 12 English and 13 Mandarin listeners participated in an experiment involving three tasks: (1) Mandarin tone identification—a clearly linguistic task where a strong effect of language background was expected, (2) pure‐tone and pulse‐train frequency discrimination—a clearly nonlinguistic auditory discrimination task where no effect of language background was expected, and (3) pitch glide identification—a nonlinguistic auditory categorization task where some effect of language background was expected. As anticipated, Mandarin listeners identified Mandarin tones significantly more accurately than English listeners (Task 1) and the two groups’ pure‐tone and pulse‐train frequency discrimination thresholds did not differ (Task 2). For pitch glide identification (Task 3), Mandarin listeners made more identification errors: in comparison with English listeners, Mandarin listeners more frequently...

This study investigated several temporal features of English to determine the extent of their occurrence in the speech of talkers of Chinese‐accented English who had relatively limited experience with spoken English. Specifically, the extent to which these speakers produced the following temporal contrasts was examined: (1) tense versus lax vowel duration, (2) vowel duration before voiced versus voiceless consonants, and (3) vowel and consonant duration in sentence‐final versus nonfinal position. Preliminary data from sentences produced by eight non‐native and eight native talkers indicates that the native English speakers and the Chinese‐accented talkers did not differ in the extent to which they realized the inherent duration difference between tense and lax vowels. However, the native English subjects tended to show substantially greater vowel lengthening before voiced versus before voiceless consonants than the Chinese‐accented talkers. In addition, while the two groups did not differ significantly in the extent to which they lengthened sentence‐final consonants relative to nonfinal consonants, the native English talkers showed greater sentence‐final vowel lengthening than the Chinese‐accented talkers. When group differences for a given temporal parameter were found, 1 or 2 of the non‐native subjects typically fell within the range of performance shown by the native speakers.

For non‐native listeners, intelligibility of non‐native speakers (NNS) can surpass intelligibility of native speakers [T. Bent, J. Acoust. Soc. Am. 109, 2472 (2001)]. The present study further investigated how language background and L2 proficiency of talker and listener affect L2 intelligibility. Two Chinese NNS, two Korean NNS, and one English monolingual were recorded reading simple English sentences. Listeners were English monolinguals, Chinese NNS, Korean NNS, and NNS with other L1s. For native English listeners, the native English talker was most intelligible. For non‐native listeners, intelligibility of non‐native talkers with high degrees of L2 proficiency was better than or equal to native talker intelligibility. This pattern held regardless of whether non‐native talkers and listeners matched in L1. Additionally, the interlanguage benefit (i.e., the intelligibility difference between non‐native and native talkers) decreased as the listeners’ proficiency in English increased. This interlanguage benefit can be accounted for by a combination of a shared systematic interlanguage (when talker and listener match in L1) and the influence of interlanguage universals (when talker and listener do not match in L1). However, both of these factors become less influential as listeners L2 proficiency develops causing a decrease in the interlanguage benefit. [Work supported by NIH‐NIDCD Grant DC 03762.]

This study considered the distribution of non‐native talker phonemic errors by position‐in‐word, and their differential effects on non‐native speech intelligibility. Based on the structure of words in the native language of the talkers (i.e., Chinese), it was hypothesized that word‐initial phonemes would be produced more accurately than word‐final phonemes. Additionally, based on models of lexical access (e.g., cohort model), it was hypothesized that phonemic errors in the word‐initial position would be more detrimental to intelligibility than phonemic errors later in the word. 15 talkers of Chinese‐accented English produced 16 simple declarative sentences. These recordings were phonetically transcribed to determine the number and type of production errors and presented to native English listeners for intelligibility testing. Results showed that, as predicted, word‐initial phonemes were produced more accurately (92% correct) than word‐final phonemes (82% correct) [t(14)=3.815, p

Coda consonant clusters of monosyllabic words were studied to determine what durational modifications occur at the segment level as a function of the number, the order, and the morphemic status of segments in the coda. We focused on the segment /s/ because it has the fewest constraints in terms of position in coda clusters, allowing reversals such as tax (/tae ks/) and task (/taesk/), and because /s/ can also be a separate morpheme, allowing us to compare segment durations for tax and tacks (/tae ks/). Preliminary results from four talkers reaffirm previous findings that segment durations decrease as the number of segments increases, although all segments do not decrease to the same extent. Our data also reaffirm previous claims that morphemic /s/ has a longer duration relative to the overall word duration than its nonmorphemic counterpart. In addition to these factors, position at the right‐edge boundary of a word seems to play a special role in determining the relative durations of segments within a coda cluster. Specifically, we found that the balance between segment durations in /sk/ and /ks/ coda clusters in monomorphemic words tended to tip in favor of the word‐final segment and against the word‐internal member of the cluster.

Non‐native listeners often have particular difficulty perceiving novel phoneme contrasts, and are disproportionately affected by background noise relative to native listeners [e.g., Mayo et al., J. Speech Hear. Res. 40, 686–693 (1997); Meador et al., Bilingualism 3, 55–67 (2000)]. This study investigated whether these difficulties could be overcome by naturally produced clear speech. Based on previous findings that hearing‐impaired adults [Picheney et al., J. Speech Hear. Res. 28, 96–103 (1985)] and learning‐impaired children [Bradlow et al., J. Acoust. Soc. Am. 108, 2603 (2000)] derive great benefit from clear speech (17%–20% and ∼9%, respectively), we predicted that non‐native listeners would also exhibit a substantial clear speech effect. The 32 non‐native and 32 native listeners were presented with English sentences. Factors that varied were speaking mode (conversational versus clear), signal‐to‐noise ratio (−4 vs −8 dB) and talker (male versus female). Contrary to our prediction, non‐native listeners...

This paper presents the results of a pilot experiment for a larger study of the effect of inventory size and structure on vowel production and perception. This study will use acoustic and perceptual data from languages which vary in the size and structure of their vowel inventories to address questions regarding the connection between vowel production and perception, and regarding the universality of language specificity of vowel space categorizations. The data for this pilot study are from four speakers of each of two languages (English and Spanish), which differ in the size of their vowel inventories: English has 11 distinct vowel qualities whereas Spanish has 5. For the production study, recordings of words which illustrate the vowel contrasts were digitized and the formant frequencies of the target vowel steady states were measured from LPC spectra. The resultant vowel charts were plotted using various scales (koenig, mel, bark, log frequency), which were then compared in terms of their efficiency in ...

This study sought to establish connections between two findings regarding speech perception in special populations. First, children with learning problems often have difficulty perceiving speech in noise. Second, naturally produced clear speech is more intelligible for adult hearing‐impaired listeners. Given recent interest in the use of digital enhancement techniques in speech and language training procedures for learning‐disabled children, we wanted to investigate whether naturally produced clear speech yields perception benefits for this population. We presented children with learning problems (n=46) and normal children (n=17) with sentences embedded in noise. Factors that varied were speaking style (plain versus clear) and signal‐to‐noise ratio (high versus low). Results indicated that: (1) children with learning problems had poorer sentence‐in‐noise perception than normal children; (2) both groups had poorer perception with increasing noise levels; and (3) both groups benefited from naturally produced clear speech. There were no significant two‐ or three‐way interactions between listener group, noise level, and speaking style. In combination with acoustic analyses of the plain and clear sentences, these findings have broad implications for our understanding of the speech perception deficits in this population, and the design of appropriate training procedures. [Work supported by NIH‐NIDCD Grants DC 03762 and DC 01510.]

While the ability to perceive novel phoneme contrasts is necessary for accurate spoken word recognition by non‐native listeners, it is likely that novel phoneme perception functions in a linguistically meaningful manner only once the contrast in question signals a known lexical contrast. In other words, knowledge of the sound‐based structure of the target language lexicon is just as important for non‐native speech perception as experience with the structure of the target language phoneme inventory. In order to understand non‐native speech perception fully, there is a need to investigate recognition of word‐sized units by non‐native listeners using stimulus materials that are well controlled in terms of the sound‐based structure of the target language lexicon. Accordingly, how non‐native listeners recognize English words under conditions known to affect spoken word recognition by native English listeners was investigated. Results showed that the ability to take advantage of consistent surface phonetic information, such as a consistent talker across items, is a perceptual skill that transfers easily from first to second language perception. However, non‐native listeners had particular difficulty with words that are easily confused with other similar sounding words, suggesting that non‐native word recognition may be compromised when fine phonetic discrimation at the segmental level is required.

Previous research has established that many children with learning problems exhibit perceptual deficits in response to certain auditory stimuli, such as stop consonants in CV context. In order to investigate the precise acoustic features of stop consonants that pose perceptual difficulties for these children, discrimination thresholds were compared for normal and learning‐disabled children along two separate synthetic /da/–/ga/ continua that differed only in the duration of the formant transitions. Results showed that simply lengthening the formant transition duration from 40 to 80 ms did not result in improved discrimination thresholds for the learning‐disabled children. Similarly, an electrophysiologic response that is known to reflect the brain’s preconscious ability to detect a change from one auditory stimulus to another indicated diminished responses in these children to /da/ versus /ga/, regardless of transition duration. This finding suggests that the brevity of the CV formant transitions is not the sole acoustic basis for the observed perceptual deficits. Furthermore, signal processing techniques aimed at enhancing perception of stop consonants likely need to consider other acoustic features, such as relative amplitude and frequency characteristics, instead of (or in addition to) CV transition duration. [Work supported by NIH R01 DC 01510.]

Various acoustic–phonetic and listener–transcription analyses were conducted using materials from two large databases of recorded speech. The first database consisted of sentence‐length materials spoken by 20 talkers, along with intelligibility data in the form of transcriptions by ten native English listeners per talker. Acoustic–phonetic analyses of the digitized speech samples showed that talkers who exhibited a high degree of ‘‘articulatory precision’’ had higher intelligibility scores than talkers who produced more ‘‘reduced’’ speech. The second database consisted of a set of words spoken by ten talkers at three speaking rates, along with transcriptions by native English listeners. Analyses of the transcription accuracy scores showed a strong effect of inherent lexical characteristics, and a strong effect of speaking rate. Furthermore, the difficulties imposed by one factor, such as a fast speaking rate or an inherently difficult lexical item, could be overcome by the advantage gained through the lis...

Two groups of Japanese speakers were trained to identify AE /r/ and /l/ using two different types of training: audio visual and auditory only. In audiovisual training, a movie of the talker’s face was presented together with the auditory stimuli, whereas only auditory stimuli were presented in audio‐only training. Improvement in /r/–/l/ identification from pretest to post‐test on three types of tests (audio‐only, visual only and audiovisual) did not differ substantially across the two training groups. Interestingly, the audio‐only group showed improved identification in the visual‐only tests, suggesting that training in the auditory domain transferred to the visual domain. A McGurk‐type test using /r/ and /l/ stimuli with conflicting audio and visual information was also conducted. Identification accuracies on this test showed a greater effect of conflicting visual information at post‐test than at pretest for the audio–visual training group, but not for the audio‐only training group, suggesting that audio‐visual training facilitated integration of auditory and visual information. Taken together, these results suggest that the internal representation of second‐language phonetic categories incorporates both auditory and visual information. Implications for theories of perceptual learning and phonological development will be discussed.

Previous work from these laboratories has shown that monolingual Japanese adults who were subjected to intensive perceptual identification training improved in both perception and production of English /r/‐/l/ minimal pairs. The present study expanded on this finding by comparing the results from three different perceptual analysis procedures. The initial test was a direct comparison by American English listeners of the Japanese trainees’ pre‐test and post‐test utterances. This test assessed the extent to which AE listeners could discriminate pre‐test and post‐test productions. The second test used a two‐alternative forced‐choice identification paradigm to assess the intelligibility of the pre‐test and post‐test /r/–/l/ words. The final test employed an open‐set transcription task in which the AE listeners simply transcribed the Japanese productions. Each of these perceptual analysis tests submitted the Japanese productions to increasingly stringent judgment criteria, and thus provided different informati...

Even within careful speaking conditions, normal speakers vary in intelligibility. This study investigated vowel space characteristics as one correlate of variation in intelligibility. Data consisted of 100 Harvard sentences, spoken by 20 speakers and transcribed by ten listeners per speaker. Formant measurements (mel transformed) of six tokens each for three vowels [i,a,o] provided a basis for assessing the vowel‐space characteristics. First, vowel‐space area was calculated as the Euclidean area between vowel category means. Across all 20 speakers, this area did not correlate with intelligibility. However, for a subset of the speakers (three highest, three lowest in intelligibility) there was a moderate positive correlation (r=+0.34). Second, vowel‐space dispersion was calculated as the mean distance from the center of the triangular space to all 18 tokens. This dispersion measure showed a moderate, positive correlation with intelligibility across all 20 speakers (r=+0.43), suggesting the importance of individual token location for overall intelligibility. Finally, minimum and maximum formant values gave a measure of F1 and F2 range covered by vowel tokens. This measure revealed that F1 range correlated positively with intelligibility (r=+0.53), whereas F2 range did not. The results will be discussed in the context of talker variability and intelligibility.

Previous research has shown a significant correlation between the perception and production of English /r/ and /l/ by Japanese speakers [Yamada et al., Proc. ICSLP94, 2023–2026 (1994)]. The present study further investigated this perception‐production link by examining the transfer of training in perception to production of a non‐native contrast. Twelve monolingual Japanese speakers were trained to perceive the English /r/–/l/ contrast using a high‐variability training program [Lively et al., 2076–2087]. Recordings were also made of the trainees’ productions of English /r/‐/l/ minimal pairs before and after the perceptual training. These pretest and post‐test recordings were then evaluated perceptually by American listeners who were presented with pairs of tokens in an A‐B test format. As expected, subjects’ accuracy on the perceptual task improved by about 16% after 45 training sessions. More importantly, subjects’ productions at pretest and at post‐test were distinguishable by American listeners. More p...

In contrast to results of training with stimuli produced by 5 talkers, Lively et al. [ 1242–1255 (1993)] reported that Japanese adults trained to perceive English /r/ and /l/ with stimuli produced by a single talker failed to improve from pretest to post‐test, or to generalize to novel stimuli. That study was extended by training 5 groups of subjects each with a different talker, and by examining the retention of learning after 3 and 6 months. The previous results were partially replicated: Although all subjects showed significant learning during training, subjects in 3 of the 5 groups did not show significant improvement in a pretest–post‐test comparison, did not generalize well to new stimuli, and did not show good retention in 3‐ and 6‐month follow‐up tests. Subjects in two of the five groups improved significantly from pretest to post‐test, generalized well to new stimuli, and showed retention comparable to that of subjects trained with multiple talkers. The results indicate that while multiple‐talker training leads to consistently good results, training with stimuli produced by only one talker may fail to promote generalization to new stimuli and talkers under certain conditions.

Intelligibility tests of a database of 2000 recorded sentences (100 sentences produced by 20 talkers), show considerable variation in intelligibility across both sentences and talkers. This paper presents preliminary results of a study designed to identify the lexical and talker characteristics that correlate with speech intelligibility scores. To identify sentence‐specific correlates of intelligibility, lexical characteristics of the words that comprise the most versus the least intelligible sentences were compared. Results show that high‐intelligibility sentences have a higher proportion of function to content words, and consequently, a higher mean word frequency than low‐intelligibility sentences. Furthermore, lexical characteristics of the most common word‐level substitution errors indicate the importance of word frequency in overall sentence intelligibility. To identify talker‐specific correlates of intelligibility, acoustic measurements of sentences that show the most inter‐speaker variability in intelligibility were compared. For example, durational characteristics of the most common segment‐level errors indicated that the relative timing between segments is an important aspect of talker intelligibility. Future directions and general implications of this database project will be discussed. [Work supported by NIH Training Grant No. T32 DC‐00012 and NIH Research Grant No. R01 DC‐00111.]

This paper addresses three issues. First, the effect of inventory size on the acoustic realization of vowels in English and Spanish is examined. Results indicate that the English vowel space differs from the Spanish vowel space in range, but not in tightness of within‐category clustering. The second experiment tests the hypothesis that the perceptual categories coincide with the produced categories within a given language. Indeed, the data suggest that the perceptual and acoustic vowel spaces within English and Spanish correspond closely to each other. Finally, this study explores cross‐language differences in vowel perception. These data suggest that there are universal, perhaps psychophysical, boundaries which are independent of inventory size: in both languages, the boundaries between /i/ and /e/, and between /u/ and /o/ are primarily in the F1 dimension, and are in the 400‐ to 500‐mels region. This suggests that the observed regularities in vowel inventories across languages may stem, in part, from universal, auditory‐perceptual mechanisms. Results of experiments in which English listeners responded to Spanish‐based stimuli, and vice‐versa, will also be discussed.

This paper reports on three experiments [J. Ohala and D. Feder, J. Acoust. Soc. Am. Suppl. 1 80, S110 (1986)], designed to show that inferencing occurs commonly in speech perception. This challenges the direct‐realist theory [C. A. Fowler, J. Phon. 14, 3–28 (1986)] that claims that speech perception is direct, and not mediated by any cognitive processes. A preliminary experiment showed that the category boundary between [i] and [u] shifts toward [u] next to a bilabial C, and toward [i] next to an alveolar. This boundary shift recurs in V1CV2 stimuli in which the C was masked by white noise. (V1 varies between [i] and [u]; V2 is always [a].) These masked stimuli were presented in blocks where the medial C was always either [b] or [d], which allowed the masked consonant to be “restored” based on the audible consonant in the unmasked stimuli. In the second experiment, 80% of the stimuli in each block had either [b] or [d], and the remaining 20% had the contrasting place. In the third experiment, the 20% minority contained [g]'s. In both cases, the vowel category boundary should exhibit an effect of the place of the 80% “context” stimuli. With [b] or [d], the place of the 20% minority should also effect the [i]‐[u] boundary, but [g] should not shift the boundary because listeners would expect [g], unlike [b] and [d], to vary with vowel backness. Preliminary analyses indicate thai expectation participates jointly with direct acoustic analysis in speech perception.

In vivo, deoxyhemoglobin S fibers associate into bundles, or fascicles. It is the formation of these bundles within red blood cells which leads to the clinical manifestations of sickle cell disease. A similar phenomenon is observed in vitro. Fascicles are initially small and poorly ordered but, with time, become progressively larger, more tightly packed and more highly ordered. The most tightly packed fascicles display crystalline order and form “crystals of fibers”. We have constructed computer generated models of longitudinal projections of crystalline fascicles of sickle hemoglobin (HbS) fibers. A comparison of the models with electron micrographs provides a means of accurately determining the three dimensional packing arrangement of fibers and the nature of the fiber-fiber interactions within fascicles.Electron micrographs of longitudinal sections of well ordered fascicles reveal a number of characteristic views, one of which is shown in figure 2a. The different views observed in the micrographs arise because of differing orientations of fascicle sections relative to the angle of projection. This was demonstrated by experiments in which the specimen was tilted about its long axis. Different views could be interconverted by varying the angle of tilt.