Search

Vowel production in gay, lesbian, bisexual (GLB), and heterosexual speakers was examined. Differences in the acoustic characteristics of vowels were found as a function of sexual orientation. Lesbian and bisexual women produced less fronted /u/ and /ɑ/ than heterosexual women. Gay men produced a more expanded vowel space than heterosexual men. However, the vowels of GLB speakers were not generally shifted toward vowel patterns typical of the opposite sex. These results are inconsistent with the conjecture that innate biological factors have a broadly feminizing influence on the speech of gay men and a broadly masculinizing influence on the speech of lesbian/bisexual women. They are consistent with the idea that innate biological factors influence GLB speech patterns indirectly by causing selective adoption of certain speech patterns characteristic of the opposite sex.

Previous work has established that naturally produced clear speech is more intelligible than conversational speech for adult hearing-impaired listeners and normal-hearing listeners under degraded listening conditions. The major goal of the present study was to investigate the extent to which naturally produced clear speech is an effective intelligibility enhancement strategy for non-native listeners. Thirty-two non-native and 32 native listeners were presented with naturally produced English sentences. Factors that varied were speaking style (conversational versus clear), signal-to-noise ratio (-4 versus -8 dB) and talker (one male versus one female). Results showed that while native listeners derived a substantial benefit from naturally produced clear speech (an improvement of about 16 rau units on a keyword-correct count), non-native listeners exhibited only a small clear speech effect (an improvement of only 5 rau units). This relatively small clear speech effect for non-native listeners is interpreted as a consequence of the fact that clear speech is essentially native-listener oriented, and therefore is only beneficial to listeners with extensive experience with the sound structure of the target language.

Northwestern University has a vibrant and highly interdisciplinary community of acousticians. Of the 13 ASA technical areas, three have strong representation at Northwestern: Speech Communication, Psychological and Physiological Acoustics, and Musical Acoustics. Sound-related work is conducted across a wide range of departments including Linguistics (in the Weinberg College of Arts and Sciences), Communication Sciences & Disorders, and Radio/Television/Film (both in the School of Communication), Electrical Engineering & Computer Science (in the McCormick School of Engineering), Music Theory & Cognition (in the Bienen School of Music), and Otolaryngology (in the Feinberg School of Medicine). In addition, The Knowles Hearing Center involves researchers and labs across the university dedicated to the prevention, diagnosis and treatment of hearing disorders. Specific acoustics research topics across the university range from speech perception and production across the lifespan and across languages, dialect an...

Switched dominance bilinguals (i.e., “heritage speakers,” HS, with L2 rather than L1 dominance) have exhibited native-like heritage language (L1) sound perception (e.g., Korean three-way VOT contrast discrimination by Korean HS; Oh, Jun, Knightly, & Au, 2003) and sound production (e.g., Spanish VOT productions by Spanish HS; Au, Knightly, Jun, & Oh, 2002), but far from native-like proficiency in other aspects of L1 function, including morphosyntax (Montrul, 2010). We investigated whether native-like L1 sound production proficiency extended to heritage language sentence-in-noise intelligibility. We recorded English and Spanish sentences by Spanish HS (SHS) and monolingual English controls (English only). Native listeners of each language transcribed these recordings under easy (-4 dB SNR) and hard (-8 dB SNR) conditions. In easy conditions, SHS English and Spanish intelligibility were not significantly different, yet in hard conditions, SHS English intelligibility was significantly higher than SHS Spanish intelligibility. Furthermore, we observed no differences between SHS English and English-control intelligibility in both conditions. These results suggest for SHS, while early Spanish exposure provided some resistance to heritage language/L1 intelligibility degradation, the absence of continuous Spanish usage impacted intelligibility in severely degraded conditions. In contrast, the absence of early English exposure was entirely overcome by later English dominance.

Speech recognition is highly sensitive to adverse conditions at all stages of the speech chain, i.e., the sequence of events that transmits a message from the mind/brain of a speaker through the acoustic medium to the mind/brain of a listener. Adverse conditions can originate from source degradations (e.g., disordered or foreign-accented speech), environmental disturbances (e.g., background sounds with or without energetic masking), and/or receiver (i.e., listener) limitations (e.g., impaired or incomplete language models, peripheral deficiencies, or tasks with high cognitive load). (For more on this classification system, see Mattys, Davis, Bradlow, & Scott, 2012, Language and Cognitive Processes, 27). This talk will present a series of studies focused on linguistic aspects of these various possible sources of adverse conditions for speech recognition. In particular, we will demonstrate separate and combined influences of the talker’s language background (a possible source degradation), the presence of a background speech masker in either the same or a different language from that of the target speech (a possible environmental degradation), and the listener’s experience with the language of the target and/or masking speech (a possible receiver limitation). Together, these studies demonstrate strong influences of language and linguistic experience on speech recognition in adverse conditions.

Speech processing can often take place in listening conditions that involve the mixing of speech and background noise. This study used a speeded classification paradigm to investigate whether background noise is perceptually integrated with indexical (Exp. 1) and phonetic (Exp. 2) dimensions of the speech signal. In each experiment, English listeners classified words along one of two dimensions: noise (pure tone vs. white noise) or a speech dimension, either gender (male vs. female in Exp. 1a), talker identity (Sue vs. Carol in Exp. 1b), or phoneme (/p/ vs. /b/ in Exp. 2), while ignoring the other dimension, which could be held constant (control), co-vary (correlated), or vary randomly (orthogonal). The results indicated that background noise was not completely segregated from speech, even when the two auditory streams were spectrally non-overlapping. Perceptual interference was asymmetric, whereby irrelevant indexical and phonetic variation slowed noise classification to a greater extent than the reverse...

The concept of subliminal perception has been a subject of interest and controversy for decades. Of interest in the present investigation was whether a neurophysiologic index of stimulus change could be elicited to speech sound contrasts that were consciously indiscriminable. The stimuli were chosen on the basis of each individual subject's discrimination threshold. The speech stimuli (which varied along an F3 onset frequency continuum from /da/ to /ga/) were synthesized so that the acoustical properties of the stimuli could be tightly controlled. Subthreshold and suprathreshold stimuli were chosen on the basis of behavioral ability demonstrated during psychophysical testing. A significant neural representation of stimulus change, reflected by the mismatch negativity response, was obtained in all but 1 subject in response to subthreshold stimuli. Grand average responses differed significantly from responses obtained in a control condition consisting of physiologic responses elicited by physically identical stimuli. Furthermore, responses to suprathreshold stimuli (close to threshold) did not differ significantly from subthreshold responses with respect to latency, amplitude, or area. These results suggest that neural representation of consciously imperceptible stimulus differences occurs and that this representation occurs at a preattentive level.

The neural representation of sensory events depends upon neural synchrony. Auditory neuropathy, a disorder of stimulus-timing-related neural synchrony, provides a model for studying the role of synchrony in auditory perception. This article presents electrophysiological and behavioral data from a rare case of auditory neuropathy in a woman with normal hearing thresholds, making it possible to separate audibility from neuropathy. The experimental results, which encompass a wide range of auditory perceptual abilities and neurophysiologic responses to sound, provide new information linking neural synchrony with auditory perception. Findings illustrate that optimal eighth nerve and auditory brainstem synchrony do not appear to be essential for understanding speech in quiet listening situations. However, synchrony is critical for understanding speech in the presence of noise.

Previous work from our laboratories has shown that monolingual Japanese adults who were given intensive high-variability perceptual training improved in both perception and production of English /r/–/l/ minimal pairs. In this study, we extended those findings by investigating the long-term retention of learning in both perception and production of this difficult non-native contrast. Results showed that 3 months after completion of the perceptual training procedure, the Japanese trainees maintained their improved levels of performance on the perceptual identification task. Furthermore, perceptual evaluations by native American English listeners of the Japanese trainees’ pretest, posttest, and 3-month follow-up speech productions showed that the trainees retained their long-term improvements in the general quality, identifiability, and overall intelligibility of their English /r/–/l/ word productions. Taken together, the results provide further support for the efficacy of high-variability laboratory speech sound training procedures, and suggest an optimistic outlook for the application of such procedures for a wide range of “special populations.”

This study investigated the encoding of the surface form of spoken words using a continuous recognition memory task. The purpose was to compare and contrast three sources of stimulus variability--talker, speaking rate, and overall amplitude--to determine the extent to which each source of variability is retained in episodic memory. In Experiment 1, listeners judged whether each word in a list of spoken words was "old" (had occurred previously in the list) or "new." Listeners were more accurate at recognizing a word as old if it was repeated by the same talker and at the same speaking rate; however, there was no recognition advantage for words repeated at the same overall amplitude. In Experiment 2, listeners were first asked to judge whether each word was old or new, as before, and then they had to explicitly judge whether it was repeated by the same talker, at the same rate, or at the same amplitude. On the first task, listeners again showed an advantage in recognition memory for words repeated by the same talker and at same speaking rate, but no advantage occurred for the amplitude condition. However, in all three conditions, listeners were able to explicitly detect whether an old word was repeated by the same talker, at the same rate, or at the same amplitude. These data suggest that although information about all three properties of spoken words is encoded and retained in memory, each source of stimulus variation differs in the extent to which it affects episodic memory for spoken words.

This study investigates the effect of hyperarticulated, intelligibility‐enhancing clear speech on English speech rhythm. Ramus et al. [Ramus et al., ‘‘Correlates of linguistic rhythm in the speech signal,’’ Cognition 72, 265–292 (1999)] showed that temporal properties of a speech signal such as the percentage of vocalic intervals (%V) and variability of consonantal and vocalic intervals (C, V) can be related to phonological properties such as presence/absence of unstressed vowel reduction and syllable structure complexity, and are consequently quite successful at grouping languages into the traditional rhythmic classes (stress‐, syllable‐, and mora‐timed). Here, we explore whether/how clear speech affects stress‐timed characteristics of English sentences in terms of these measures. Results revealed that the proportion of vocalic intervals (%V) within sentences remained stable across speaking styles, i.e., consonants and vowels were lengthened equally in clear speech. However, variability of both vocalic and consonantal intervals (V, C) was higher in clear than in conversational speech. The increase in V and C was achieved primarily through insertion/strengthening of short vocalic and consonantal segments that were dropped or coarticulated with surrounding sounds in conversational speech, making word and syllable boundaries more salient. These results suggest that increased intelligibility of clear speech can in part be attributed to prosodic structure enhancement by means of enhanced syllable and word boundaries demarcation.This study investigates the effect of hyperarticulated, intelligibility‐enhancing clear speech on English speech rhythm. Ramus et al. [Ramus et al., ‘‘Correlates of linguistic rhythm in the speech signal,’’ Cognition 72, 265–292 (1999)] showed that temporal properties of a speech signal such as the percentage of vocalic intervals (%V) and variability of consonantal and vocalic intervals (C, V) can be related to phonological properties such as presence/absence of unstressed vowel reduction and syllable structure complexity, and are consequently quite successful at grouping languages into the traditional rhythmic classes (stress‐, syllable‐, and mora‐timed). Here, we explore whether/how clear speech affects stress‐timed characteristics of English sentences in terms of these measures. Results revealed that the proportion of vocalic intervals (%V) within sentences remained stable across speaking styles, i.e., consonants and vowels were lengthened equally in clear speech. However, variability of both vocalic a...

When a firm conducts business across borders to exploit location, scale or scope economies, it inevitably encounters nontrivial costs of language communication that can dissipate the value that it ...

This study investigated the effects of training in /r/–/l/ perceptual identification on /r/–/l/ production by adult Japanese speakers. Subjects were recorded producing English words that contrast /r/ and /l/ before and after participating in an extended period of /r/–/l/ identification training using a high-variability presentation format. All subjects showed significant perceptual learning as a result of the training program, and this perceptual learning generalized to novel items spoken by new talkers. Improvement in the Japanese trainees’ /r/–/l/ spoken utterances as a consequence of perceptual training was evaluated using two separate tests with native English listeners. First, a direct comparison of the pretest and post-test tokens showed significant improvement in the perceived rating of /r/ and /l/ productions as a consequence of perceptual learning. Second, the post-test productions were more accurately identified by English listeners than the pretest productions in a two-alternative minimal-pair identification procedure. These results indicate that the knowledge gained during perceptual learning of /r/ and /l/ transferred to the production domain, and thus provides novel information regarding the relationship between speech perception and production.

Previous research suggests that accuracy (i.e., distance to the average location of native productions) has less effect on adaption to non-native speech than category variability [e.g., Wade et al., Phonetica 64, 122-144 (2007)]. Here we investigate the relationship between overall intelligibility of Mandarin-accented English for native English listeners and (a) vowel production accuracy, and (b) vowel production consistency. Intelligibility estimates were based on sentence-in-noise recognition accuracy scores. Vowel accuracy and consistency estimates were based on formant measurements of point vowels (/i/, /u/, /ae/, and /a/) extracted from words in the sentence materials that were presented to listeners for intelligibility testing (8-20 samples/vowel/talker). If listeners have expectations about a vowel category location based on accumulated exemplar storage, then greater accuracy (smaller Euclidean distance to native category mean) should be positively related to intelligibility. If listeners are sensitive to vowel category distributions, then greater consistency (smaller standard deviation of f1 or f2 within categories) should be beneficial to intelligibility. A mixed effects linear model revealed that only consistency was a significant predictor of intelligibility. Accuracy was not a significant predictor. This result suggests that intra-speaker variability is detrimental to L2 intelligibility, regardless of distance to native categories.

Talker familiarity can facilitate the extraction of linguistic content from speech signals embedded in broadband noise; however, relatively little research has investigated the impact of talker familiarity with competing speech in the background. This study explores the effects of familiarity with the target or competing talker in speech-in-speech perception. Listeners were first familiarized with and trained to identify three female voices. They then completed a sentence recognition task in the presence of 1-talker babble. Familiarity with either the target or background talker was manipulated in separate conditions. Results revealed significantly better sentence recognition for familiar relative to unfamiliar target talkers in the presence of an unfamiliar background talker; however, sentence recognition with an unfamiliar target talker did not differ depending on background talker familiarity. Thus, while listeners were able to capitalize on familiarity with a talker’s voice to aid target speech recognition, familiarity with the competing talker was neither facilitative nor inhibitory. This suggests that the influence of talker familiarity is limited to the actively attended stream. This stands in contrast to other aspects of the unattended stream which have been shown to exert an influence on speech-in-speech recognition including language and semantic content of the background speech.

Nonnative talkers tend to exhibit slower speech rates than native talkers at the group level. Here we ask whether individual variation in rate is language-general to the extent that L1 rate is a significant predictor of L2 rate within bilinguals. 62 nonnative English talkers participated in three speech production tasks in both their L1 (14 Cantonese, 14 Mandarin, 11 Korean, 4 Portuguese-Brazilian, 6 Spanish, 13 Turkish) and L2 (English), namely, reading a paragraph, spontaneously answering questions, and spontaneously describing a picture story. Two measurements of rate were automatically extracted from the recordings: speech rate (syllables per second), and articulation rate (syllables per second excluding silent pauses). As expected, L2 speech and articulation rates were overall slower than L1 speech and articulation rates for all tasks. Importantly, L2 speech rates and articulation rates were positively related to L1 speech rates and articulation rates, respectively. There were also significant differences in L2 speech rates and L2 articulation rates depending on L1 background and tasks. However, the positive relationship between L1 and L2 rates still holds with these other effects taken into consideration, suggesting that overall rate variation is partially an individual-specific property that transcends L1 and L2 within bilinguals. Acknowledgments: Vanessa Dopker and Chun Liang Chan. [Work supported by Grant R01-DC005794 from NIH-NIDCD.]

Many English conversations across the globe today involve talkers with different language experiences. Here we show that, while language barriers challenge communicative efficiency, the detrimental effect of language distance may be mitigated by phonetic convergence. We analyzed a corpus of conversations in which talker pairs solved a spot-the-difference puzzle by verbally comparing two scenes only one of which was visible to each talker ("diapix" task). Language distance was varied by pairing talkers who either matched or mismatched in language background and in native/nonnative status relative to the target language. Communicative efficiency was measured by task-completion-time and word type-to-token ratio. Phonetic convergence was assessed by perceptual similarity tests in which listeners compared samples from one talker's speech to samples from his/her partner's speech from either early or late portions of the conversation. In this test, greater similarity for late than early samples indicates convergence. Results showed a negative correlation between language distance and communicative efficiency, a negative correlation between language distance and phonetic convergence, and a mitigating effect of phonetic convergence on the negative correlation between language distance and communicative efficiency. This suggests that convergence may be an effective mechanism for overcoming the detrimental effects of a language barrier.

To examine the relationship between speech perception and production in second language acquisition, this study investigated whether training in the perception domain transfers to improvement in the production domain. Native speakers of Japanese were trained to identify English /r/-/l/ minimal pairs. Recordings were made of the subjects' productions of minimal pairs before and after identification training. American-English listeners then perceptually evaluated these productions. The subjects showed significant improvements from pre-test to post-test in perception as well as in production. Furthermore, the subjects retained these abilities in follow-up tests given three months and six months after the conclusion of training. These results demonstrate that training in the perception domain produces long-term modifications in both perception and production, implying a close link between speech perception and production.

When a firm conducts business across borders to exploit location, scale or scope economies, it inevitably encounters nontrivial costs of language communication that can dissipate the value that it ...

Languages differ widely in the size of their vowel inventories; however, cross‐linguistic surveys indicate that certain vowels and vowel system configurations are preferred. A cross‐linguistic comparison of the acoustic vowel categories of two languages that differ in vowel inventory size, namely, English and Spanish, was performed in order to reveal some of the language‐specific and/or universal principles that determine the acoustic realization of the vowels of these two languages. This comparison shows that the precise location in the acoustic space of similar vowel categories across the two languages is determined, in part, by a language‐specific base‐of‐articulation property. These data also suggest that the relatively crowded acoustic vowel space of English may be expanded with respect to the relatively uncrowded acoustic vowel space of Spanish; however, this effect is variable depending on the syllable context of the English vowels. Finally, the data indicate no difference in the tightness of within‐category clustering for the large versus the small vowel inventory.

We investigated native English talkers’ phonetic accommodation to a native or nonnative model talker in a passive auditory exposure setting. We performed a phonetic accommodation experiment, following the procedure of Goldinger & Azuma (2004). Specifically, the imitators read monosyllabic words, disyllabic words, and sentences before and after perceptual exposure to the stimuli. We found evidence of phonetic convergence both to native and nonnative model talkers from various acoustic measurements on words and sentences, and dynamic time warping analyses and XAB perception tests on sentences. We also found that dialect mismatch between participants and native model talkers inhibited phonetic convergence in some acoustic measurements. Additionally, the distances between model talkers and participants along the acoustic measurements before auditory exposure positively affected their degrees of phonetic convergence, regardless of the direction of the change; the farther the acoustic distance was before the auditory exposure, the larger the degree of phonetic convergence was. Moreover, the imitators generalized their accommodation patterns from exposed to unexposed items. Finally, XAB perception tests with the sentences revealed that imitators of all model talkers were perceived as converging towards their model talker, and importantly, this pattern of perceived accommodation was predicted by most of the sentence-based acoustic measurements.

This study asks whether speech recognition by bilingual listeners in each of their two languages follow complementary or supplementary patterns. Previous studies showed that early bilinguals are disproportionately affected by adverse listening conditions in L2 (Mayo et al., 1997; Shi et al, 2010; Bradlow & Alexander, 2007), but did not measure performance in L1. The current study extends these results using bilingual performance under adverse listening conditions in both languages to determine whether reduced use of the dominant language by relatively well-balanced bilinguals affects performance in L1 as well as L2. We examine two groups of English-Spanish bilinguals: Spanish learners (SL) and Spanish heritage speakers (SHS). Although both English dominant, crucial differences between these groups are L1 (SL=English, SHS=Spanish) and L1-L2 balance (SL=large imbalance, SHS=relatively balanced). Both groups were presented with sentences in English and Spanish in which final keywords varied on three factors: speech style (clear versus plain/conversational), contextual predictability (high versus low), and signal-to-noise ratio (easier versus harder). Results show SHS do not pattern like monolinguals in either language, yet average performance across both languages is higher than SL. This result suggests the overall system SHS maintain is “larger” than SL, but may be more susceptible to noise.

This study investigated variability in L2 speech intelligibility as a function of L1 speech intelligibility and of talker-listener L1 match. Non-native Korean talkers varying in their L2 proficiency were recorded reading simple English (L2) and Korean (L1) sentences. The intelligibility of these sentences was then assessed by Korean listeners (both Korean and English productions) and English listeners (English productions only) in a sentence recognition task. The results revealed that for these Korean-English bilingual talkers, L1 intelligibility was significantly correlated with L2 intelligibility for both Korean and English listeners, suggesting that variability in L1 speech intelligibility can serve as a predictor of variability in L2 production accuracy. We also examined the interlanguage speech intelligibility benefit for non-native listeners (ISIB-L) (e.g., Bent & Bradlow, 2003; Hayes-Harb et al., 2008). Korean listeners performed better at identifying the sentences produced by Korean talkers with relatively low L2 intelligibility, implying that the benefit of a shared native language between talkers and listeners may be larger when non-native listeners process speech from a talker with low L2 intelligibility. Overall, these findings indicate that variability in L2 speech intelligibility is related to language-general talker characteristics as well as to the talker-listener language alignment.

Previous research has shown that L1 speech intelligibility, as judged by native listeners, varies due to speaker-specific characteristics. Similarly, L2 speech intelligibility as judged by native listeners also varies across speakers. Given variability in L1 and L2 intelligibility, we hypothesize that, within bilinguals, some speaker-specific characteristics that contribute to variability in L1 intelligibility (e.g., long-term average spectrum, speech rate, and articulatory precision) are language-independent and therefore also contribute to variability in L2 intelligibility. This leads to the expectation that within a group of bilingual speakers, relative L1 intelligibility is a significant predictor of relative L2 intelligibility. In the current study, 14 Mandarin-English bilinguals produced 112 short meaningful sentences in their L1 (Mandarin) and L2 (English). Independent groups of Mandarin and English listeners then repeated back the sentences (native-accented Mandarin productions for Mandarin listeners at -4 dB SNR, Mandarin-accented English productionsfor English listeners at 0 dB SNR). Intelligibility was calculated as proportion of words correctly repeated. L1 intelligibility was a significant predictor of L2 intelligibility (β=.58, S.E.β=.21, p

Clear speech is an intelligibility-enhancing mode of communication often used when speakers have trouble being understood. Previous work has established that both native and non-native listeners can receive a clear speech perception benefit, though possibly to differing degrees (Bradlow and Bent, 2002). Few studies have looked at whether nonnative talkers can induce this clear speech benefit (e.g., Smiljanic and Bradlow, 2007; Rogers etal., 2010). The current study examined English clear and conversational speech by nonnative speakers from three language backgrounds (Japanese, Portuguese, and Turkish) and two proficiency levels to determine their effect on the inducement of a clear speech benefit. Native English listeners repeated back semantically anomalous sentences. The signal to noise ratio was adjusted to the level at which they could correctly repeat 50% of the words using an adaptive test similar to the Hearing in Noise Test (Nilsson etal., 1993). The results suggest that the speaker's native language may play a role in the size of the induced clear speech benefit independently of proficiency level. Additionally, accent ratings indicated a dissociation of intelligibility and accentedness. These results are consistent with the notion that variability in intelligibility is subject to language-specific knowledge by both the talker and the listener.

Previous research has shown that monolingual English listeners receive a release from informational masking if the competing speech is foreign‐language versus native‐language noise [e.g., Van Engen & Bradlow (2007)]. This study examines whether speech‐in‐speech recognition varies even across typologically close target and noise languages (English versus Dutch) as well as across variation in the semantic content of the masker (meaningful versus semantically anomalous sentences) and the listener status (native versus nonnative listeners). The English and Dutch listeners’ task was to repeat target sentences in noise. The results for the Dutch listeners on Dutch targets showed a release from masking in English versus Dutch noise and in semantically anomalous versus meaningful sentences. Unexpectedly, the results for both listener groups on English targets only showed a release from masking for English anomalous noise relative to all other conditions. However, after normalization of the long term average spectrum of all noise, the results with English targets also showed a release from masking in foreign‐language noise and in semantically anomalous sentences. Taken together, these speech‐in‐speech recognition studies show that linguistic masking involves interference at the acoustic‐phonetic and the semantic levels. Moreover, the listener’s native or nonnative status influences the size of the release from masking.

The context surrounding words can facilitate word recognition for native and non‐native listeners. However, for degraded speech signals, (e.g., with noise), non‐natives may lose the ability to take advantage of contextual cues unless the speech is clearly produced [Bradlow and Alexander (2007)]. This study investigated whether: (1) foreign‐accented speech degrades the signal such that natives show reduced ability to take advantage of context, and (2) both native and non‐native listeners adapt to foreign‐accented speech. Native and non‐native listeners were exposed to accented speech in two blocks. In each block, half the sentence final words were in high and half were in low predictability contexts. Listeners were asked to identify sentence final words. In a single talker condition, natives benefited from context and adapted to the accent; non‐natives could not use context but showed adaption. In a multitalker condition, natives neither benefited from context nor adapted to the accent. Non‐natives benefit...

Previous work established that both native and non‐native clear speech increase intelligibility for native and proficient non‐native listeners. However, non‐native clear speech enhanced intelligibility less than native clear speech. In this study, we examine native and non‐native conversational and clear speech productions with an eye on the differences in hyperarticulation strategies that may underlie a smaller clear speech intelligibility benefit. Results revealed a slower conversational speaking rate for non‐native talkers compared with native talkers. Clear speaking rate was similar for both groups suggesting a limit in the speaking rate decrease in intelligibility enhancing clear speech. The durational distinction between tense and lax vowels was increased less in non‐native speech due to the smaller speaking rate decrease in clear speech. Analyses of the stop voice onset time (VOT) showed that non‐native talkers lengthened the voiced category in the negative VOT direction, while the voiceless catego...

Normal‐hearing monolingual listeners whose native language is English experience a release from masking when the competing speech stimuli are spoken in a language other than English [e.g., K. J. Van Engen and A. R. Bradlow, J. Acoust. Soc. Am. 121, 519 (2007)]. It is unclear whether this release is due to the fact that the masking speech is not spectrally matched to the target speech (energetic influence), or that listeners are unable to understand the masking speech and therefore find it less “distracting” when spoken in a non‐native language (informational influence). This study investigates listeners’ recognition of English speech presented in the presence of a continuum of five two‐talker babble maskers. The two‐talker maskers (all created using male voices) include English babble, Mandarin‐accented English babble with high intelligibility, Mandarin‐accented English babble with moderate intelligibility, Mandarin‐accented English babble with low intelligibility, and Mandarin babble. We hypothesize that...

Foreign‐accented speech presents an interesting challenge for native listeners due to its deviations from native accented speech along multiple acoustic‐phonetic dimensions. Nevertheless, since these deviations arise primarily from interactions of the native (L1) and target (L2) language sound systems, they are highly systematic both within individuals and across talkers from the same L1. Moreover, since numerous features of English are typologically rare (e.g., coda clusters and large vowel inventory), various foreign‐accents of English frequently converge on similar specific deviations (e.g., cluster reduction and vowel category conflation). Thus, despite a high degree of deviation from native talker norms, the systematicity of foreign‐accented speech should facilitate listener adaptation. We examined adaptation at the level of the individual (talker‐dependent), the L1 (talker‐independent), and the population (accent‐independent). First, we showed variation in the speed and extent of talker‐dependent adaptation depending on the talker’s baseline English speech intelligibility. Second, we demonstrated that talker‐independent adaptation required exposure to multiple talkers of the accent. Finally, we demonstrated that exposure to multiple accents (e.g., Korean‐, Hindi‐, Mandarin‐, and Romanian‐accented English) can facilitate accent‐independent adaptation, i.e., adaptation to both a trained accented (Mandarin‐accented English) and an untrained accent(Slovakian‐accented English). This work builds a strong case for highly flexible speech perception mechanisms.

Speech and music both utilize pitch variation to convey contrasts in meaning: music‐pitch expresses composition (e.g., the key of a piece) and effect; speech‐pitch conveys pragmatic meaning and, in tone languages, lexical information. This study investigated whether experience with processing lexical pitch affects music pitch processing. 28 nonmusicians (14 native English speakers and 14 native Mandarin speakers) discriminated (experiment 1) and identified (experiment 2) short melodies. The Mandarin listeners more accurately discriminated the melodies than the English listeners (Mann‐Whitney U=140.5, p

The speech of sleep‐deprived persons has been variously described as more “slurred” or “tired‐sounding” than speech by the same speakers in a rested condition. Although sleep deprivation is a major focus of research for public safety and health reasons, there has been notably little research on the connection between sleep deprivation and speech. Using an automatic system for detecting “landmarks”, i.e., important perceptual and articulatory events in the acoustic signal, we compared sleep‐deprived and rested speech recorded during two well‐controlled studies of sleep deprivation. Our results show that sleep‐deprived speech is significantly different from rested speech for both the number and pattern of landmarks detected. These results are similar in detail to differences previously found between clear and conversational styles of speech. The relevance of these findings for speech as a marker of sleep‐deprivation will be discussed, along with potential applications in public safety and health. [Research supported by NIH.]

This study investigated how native language background (L1) interacts with speaking style in determining levels of speech intelligibility. In four experiments, we explored whether native and non‐native hyper‐articulation clear speech strategies provide similar intelligibility benefits for native and high proficiency non‐native listeners. The sentence‐in‐noise perception results revealed that native speech was preferred over non‐native speech by both listener groups even when non‐native talkers and listeners shared the same L1. Clear speech was shown to be beneficial for both the native and fluent non‐native listeners. However, non‐native clear speech enhanced intelligibility less than native clear speech, supporting the hypothesis that clear speech production strategies involve enhancement of language‐specific phonological contrasts. In order to assess the relationship between objective intelligibility measures and subjective accentedness ratings, we obtained accentedness ratings of native and non‐native conversational and clear speech by native and non‐native listeners. The results showed that objective intelligibility and subjective accentedness were independent. Overall, these results provide strong evidence that clear speech involves language‐specific modifications. Nevertheless, native and high proficiency non‐native clear speech modifications are generally helpful for both native and high proficiency non‐native listeners even when the objective intelligibility and subjective accentedness levels diverged for various listener and talker groups.

Many English conversations across the globe today involve non‐native speakers. To understand the sound structure of English in a global context, we asked native and non‐native listeners to rate the similarity of various languages (n=17) to English. In general, the native and non‐native listeners' distance‐from‐English ratings were positively correlated; but, the non‐natives tended to rank languages as more distant from English than the natives. Native and non‐native listeners then rated the accents of native and non‐native speakers. Again, the native and non‐native listener ratings were positively correlated, but the non‐native listeners generally rated both native and non‐native speakers as more foreign‐accented than did the native listeners. Together, these data suggest that both native and non‐native English generally sound less “marked” to native listeners than to non‐native listeners. As a means of understanding how English functions as a global language, we examined communicative efficiency in task‐oriented dialogues between various combinations of natives and non‐natives. All pairs successfully completed the task; yet, when one or more non‐native talkers were involved, the task took about twice as long to complete as when both were native talkers. Together, these studies help characterize the structure and function of English as a twenty‐first century lingua franca.

This study investigated the effect of hyperarticulated, intelligibility‐enhancing clear speech on global temporal characteristics in paragraph‐length utterances. The results of sentence‐in‐noise listening tests showed a consistent clear speech intelligibility gain across the utterances indicating that the talkers successfully maintained clear speech articulatory modifications throughout the paragraphs. We then explored temporal properties of the speech signal in terms of the percentage and variability of vocalic and consonantal intervals (following Ramus et al., 1999). Consonantal and vocalic intervals were lengthened equally in clear speech: %V remained stable across speaking styles. Moreover, coefficients of variation for both consonantal and vocalic intervals remained stable across clear and conversational speech, suggesting that global temporal properties remain rather stable in the two speaking styles. There was also evidence of an increase in the number of prosodic phrases and of vocalic and consona...

For a wide range of socio‐political reasons, many conversations across the globe today are between interlocutors who do not share a mother tongue. As a resource for investigating the nature and broad implication of speech communication in a global context, we are currently developing a large database of speech recordings from native and non‐native speakers of English. A key feature of this database is that, in addition to providing recordings of scripted materials, the speakers are recorded in pairs (all possible pairing of native and non‐native English speakers) as they work together on a novel, interactive, goal‐oriented task. The final corpus will include fully segmented and phonetically aligned digital recordings of both the scripted and unscripted speech samples along with complete orthographic transcriptions. We are currently using a first version of this database to track speaker‐listener alignment over the course of a conversation, to compare phonetic features of speech addressed to native versus to non‐native speakers, and to assess communication efficiency across various native and non‐native speaker pairings (e.g., how long does it take for a team‐based task to reach successful completion when one, both, or neither of the team members is a non‐native speaker?)

Recent speech research has established that, for humans, speech‐in‐noise and speech‐in‐quiet perception differ along numerous dimensions that span levels of signal encoding and linguistic representation. Listeners place more or less weight on specific acoustic cues, draw more or less on signal‐independent, contextual information, and are more or less distracted by lexical neighbors depending on masker type and level. Moreover, noise has different effects at different levels for different listener populations. Machine recognition of noisy speech is well below (at least by an order of magnitude) human recognition of noisy speech. This difference is true for both additive noise and convolutive noise. Researchers have focused on the problem of machine recognition of noisy speech in several ways: Developing robust features, training systems on noisy speech, or developing speech enhancement algorithms to clean up the noisy speech signal before performing recognition. A remaining challenge for understanding how humans do and how computers should handle speech‐in‐noise is to develop a conceptual framework that goes beyond the division of masking effects into peripheral, energetic versus central, informational.

In this talk, we discuss vowel production in clear speech, a distinct mode of speech production intended to enhance intelligibility, across languages with different phonological properties (Croatian and Spanish with small vowel inventories vs. English with a large vowel inventory). A comparison of Croatian and English revealed the equivalent clear speech vowel space expansion in both languages. In addition, listeners recognized words in noise more accurately in clear than in plain speech in their native language, establishing that the plain‐to‐clear speech articulatory modifications increased intelligibility. A comparison of Spanish and English also showed similar amounts of vowel space expansion as well as the maintenance of coarticulatory patterns in clear speech for both languages. The inventory‐independent vowel space expansion in all three languages suggests that talkers hyperarticulate even when segments are unlikely to be perceptually confusable (few vowel categories that are fairly distinct). The maintenance of coarticulation further suggests that spreading of segment identity across neighboring sounds may be beneficial to the listener. In this talk, we also discuss various measures of vowel space expansion and different clear speech strategies across talkers. We conclude with some remarks about the link between the identified acoustic‐phonetic features of clear speech and intelligibility.

Phonetic reduction of the second relative to the first mention of a word in a discourse (Second Mention Reduction, or SMR) is a well‐documented feature of spoken American English (AE). This phenomenon could be mediated by discourse‐level prosodic structure: second mentions are reduced because they are less likely than first mentions to be accented in AE; or it could occur independently of prosodic structure. Impressionistic studies indicate that Indian English (IE) speakers do not deaccent second mentions like AE speakers [J. Gumperz, Discourse Strategies (1982)]. Korean marks information structure through phrasing and word order rather than pitch accents. Therefore, SMR in these languages would support a direct link between lexical probability and duration. Six IE speakers read five paragraphs containing 59 repeated words, and five Korean speakers read a Korean translation of one of these paragraphs. IE speakers were significantly less likely to deaccent second mentions than AE speakers (sign‐test, p

Improvement on most perceptual tasks requires active performance of the target task during training. However, there has been little consideration of whether that active performance is necessary throughout the entire practice period. Suggesting that it is not, there is recent evidence that active performance on a simple frequency‐discrimination task, interleaved with passive stimulus exposures, can yield more learning than the active performance alone. The current study examines whether these findings extend to a speech‐learning task, specifically foreign‐accent adaptation. Native‐English speakers transcribed sentences spoken in English by non‐native speakers. Participants who performed the target task for half the training period and received passive stimulus exposures while performing a distracter task for the other half out‐performed participants who received the same amount of active training, but no passive exposures during the distracter task, suggesting that passive exposures enhanced learning. The active‐passive group also out‐performed participants who received only passive exposures—and performed as well as a group who performed only the target task—for the entire training period. Taken together, these results suggest that the combination of active training and passive stimulus exposure may play a key role in real‐world perceptual learning for speech. [Work supported by NIH/NIDCD.]

Studies examining effects of talkers’ and listeners’ linguistic backgrounds on speech intelligibility have produced mixed results, and relatively little is known about which acoustic‐phonetic features of native and non‐native speech are responsible for many findings. These issues are addressed here via investigation of how acoustic properties of final consonant voicing contrasts influence word intelligibility. Native English and native Mandarin talkers’ productions of English words were presented in isolation to native English and native Mandarin listeners in an identification task. Intelligibility findings were related to preceding vowel duration, stop consonant duration, burst duration, and proportion of voicing during stop closure. As expected, native English listeners were more accurate identifying words spoken by native than non‐native talkers. In contrast to previous findings of an interlanguage speech intelligibility benefit (ISIB) for sentences, non‐native listeners found native speech more intelligible than non‐native speech. However, non‐native listeners found non‐native speech more intelligible than did native listeners. For native talkers, all four acoustic measures contributed significantly to correct identification of voicing contrasts by both native and non‐native listeners (with closure voicing most salient). In contrast, for non‐native talkers, listeners tended to rely primarily on closure voicing and burst duration. Ramifications for ISIB will be discussed.

Previous research showed that phonological neighborhood density and word frequency influence word recognition (Luce and Pisoni, 1998) and vowel production (Wright, 2002; Munson and Solomon, 2004; Munson, to appear), suggesting an interaction of lexical and phonetic factors in speech production and perception Here, we explore whether hyperarticulated, intelligibility‐enhancing clear speech shows similar sensitivity to lexical‐level structure. Nine American English talkers (five females, four males) produced 40 monosyllabic easy (frequent words with few lexical neighbors) and hard (infrequent words with many lexical neighbors) words in conversational and clear speech. Twenty‐four subjects participated in a word‐in‐noise listening test. Results revealed a large effect of style on intelligibility and vowel production: words were more intelligible and vowels were longer and more dispersed in clear compared to conversational speech. Moreover, the female talkers produced larger vowel spaces than male talkers in both speaking styles. Vowels in hard words were marginally more dispersed than vowels in easy words in both speaking styles. However, within both speaking styles, easy and hard words were equally intelligible and of approximately equal duration. These results showed that phonetic properties of vowels were enhanced equally in clear speech regardless of their lexical properties.

Recent research has shown that listeners can classify talkers based on their regional dialect of American English (AE). The present experiment explored the effects of regional dialect variation on speech intelligibility. Participants from various AE dialects transcribed meaningful English sentences mixed with noise. The sentences were produced by male and female talkers from four different AE regional dialects. At a favorable signal‐to‐noise ratio (−2 dB), intelligibility of General American and Southern talkers was significantly better than Northern and Mid‐Atlantic talkers, and was significantly better for the Northern than the Mid‐Atlantic talkers. At a more difficult signal‐to‐noise ratio (−6 dB), intelligibility was equally poor for the Southern, Northern, and Mid‐Atlantic talkers, but remained significantly better for the General American talkers. No effects of listener dialect were observed. Thus, speech intelligibility in noise was generally poorer for more phonologically marked dialects (e.g., North, Mid‐Atlantic) than less‐marked dialects (e.g., General American). However, noise had differential effects on dialect interference; intelligibility of marked dialects was graded at easier signal‐to‐noise ratios, but marked dialects were equally difficult to transcribe at unfavorable signal‐to‐noise ratios, suggesting that dialect information may be conveyed by aspects of the signal that are relatively vulnerable to perceptual disruption by noise.

Speakers appear to adjust speech production according to tradeoffs between intelligibility and economy of effort [Lindblom (1992)]. Recently, there has been much interest in investigation of differences between the clear style of speech addressed to disadvantaged listeners (non‐native speakers, hearing impaired listeners, etc.) and ordinary, or conversational speaking style. Clear speech has been shown to be more intelligible across a wide range of listener types [Bradlow et al., (2002); Bradlow et al., (2003)], but the full range of parameters of variation remain undetermined. Recently also, the use of abrupt changes in the speech signal, i.e., acoustic landmarks, as an organizing principle for speech recognition has garnered attention [Espy‐Wilson (2005)]. Using a landmark analysis procedure based on that of Stevens (1991) and Liu (1995), we present evidence that clear and conversational speaking styles can be distinguished in terms of the distribution of particular clusters of landmarks, corresponding very roughly to syllable‐sized units. The implications of this differential distribution of landmarks across speaking styles for the organization of speech production will be discussed.

This study investigates the effect of intelligibility‐enhancing clear speech on voice onset time (VOT) in two languages with different phonetic realizations of a two‐way voicing contrast. Specifically, it was explored whether the contrasts between prevoiced and short‐lag categories in Croatian and short‐lag and long‐lag categories in English are enhanced along the VOT dimension in clear speech. Word‐initial stops in words embedded in sentences read by native speakers of each language in conversational and clear speaking styles were examined. The results revealed that in English, the long‐lag category (i.e., voiceless) extended somewhat in the positive VOT direction (longer aspiration), and in Croatian, the prevoiced category (i.e., voiced) extended somewhat in the negative VOT direction (longer prevoicing) in clear speech. However, in both languages, the short‐lag tokens (i.e., voiced in English, voiceless in Croatian) were unchanged in clear versus conversational speech. Overall, then, the distance between the two voicing categories was rather stable across the two speaking styles in both languages. These results suggest that VOT is not a dimension of voicing category contrast enhancement. Rather, the language‐specific pronunciation norms along this dimension are maintained in clear and conversational speech.

Studies on speech perception in multitalker babble have revealed asymmetries in the effects of noise on native versus foreign‐accented speech intelligibility for native listeners [Rogers et al., Lang Speech 47(2), 139–154 (2004)] and on sentence‐in‐noise perception by native versus non‐native listeners [Mayo et al., J. Speech Lang. Hear. Res., 40, 686–693 (1997)], suggesting that the linguistic backgrounds of talkers and listeners contribute to the effects of noise on speech perception. However, little attention has been paid to the language of the babble. This study tested whether the language of the noise also has asymmetrical effects on listeners. Replicating previous findings [e.g., Bronkhorst and Plomp, J. Acoust. Soc. Am., 92, 3132–3139 (1992)], the results showed poorer English sentence recognition by native English listeners in six‐talker babble than in two‐talker babble regardless of the language of the babble, demonstrating the effect of increased psychoacoustic/energetic masking. In addition, t...

The overall intelligibility advantage for sentences produced in clear versus conversational speech is well‐documented. This study looked at recognition accuracy across words in early and late positions in semantically anomalous and meaningful sentences spoken in clear versus conversational speaking styles. For both sentence types, the results showed the expected overall intelligibility advantage for clear speech over conversational speech. For the semantically anomalous sentences, in both speaking styles, a decline in keyword identification rate was observed with words earlier in the sentence being more accurately recognized than words later in the sentence. Furthermore, the intelligibility advantage for clear over conversational speech remained relatively constant across word positions. For the meaningful sentences, the decline in keyword identification rate across word positions was observed for conversational speech only. Meaningful sentences spoken in clear speech yielded a high, relatively stable word identification rate across position‐in‐sentence, resulting in a larger clear speech intelligibility benefit for words late in the sentence than for words early in the sentence. These results suggest that for typical meaningful sentences, the acoustic‐phonetic enhancements of clear speech and the availability of semantic‐contextual information combine to ‘‘boost’’ the intelligibility of words in late sentence positions.

Why is foreign‐accented speech hard to understand? In one view, foreign‐accented speech intelligibility is poor due to its marked deviation from the native talker norm. An alternative view claims that if the talkers and listeners shared knowledge systems are sufficiently aligned with each other, foreign‐accented speech intelligibility can be good despite its deviation from the native talker norm. Results from two studies in support of this latter alternative will be presented. Study 1 investigated whether intelligibility of Chinese‐accented English could be improved for native English listeners through perceptual training. This study showed that if given exposure to multiple talkers of Chinese‐accented English during training, native English listeners could achieve talker‐independent adaptation to Chinese‐accented English. Study 2 compared intelligibility of foreign‐accented and native‐accented English for non‐native and native listeners. Results showed that native listeners found native‐accented English more intelligible than foreign‐accented English, but non‐native listeners found foreign‐accented English as intelligible as native‐accented English. Interestingly, this pattern was obtained even for non‐native listeners and talkers from different language backgrounds. Taken together, these findings demonstrate that intelligibility of foreign‐accented speech varies depending on the degree of talker–listener attunement rather than on the extent of its deviation from a native talker norm per se.

This paper examines the interaction of language‐general, signal enhancement strategies and language‐specific, phonological enhancement strategies in clear speech production and perception in Croatian, a language with a phonemic vowel length contrast and a relatively small inventory of five vowel qualities. Two native speakers of Croatian (one male, one female) read 20 nonsense sentences in conversational and clear speech. In a sentence‐in‐noise perception test, native Croatian listeners more accurately recognized words produced in clear than in conversational speech by both talkers. However, this clear speech intelligibility benefit was greater for the male than the female talker. Acoustic analyses showed that in clear speech both talkers enhanced the overall acoustic salience of the signal (slower, more frequent pauses, wider pitch range) and expanded the vowel space, although the male showed slightly less extensive vowel space expansion. However, only the male talker enhanced the phonemic vowel length c...