Conversational Speaking Styles Research Articles

Protective face masks affect clearly produced diphthongs by L1 and L2 English talkers

The study investigated how the use of protective face masks and language experience shape the production of listener-oriented clear speech. One L1 and one L2 English talker read sentences in a clear and conversational speaking style with and without a surgical mask. Formant trajectories between the onset and offset of two diphthongs, /aɪ/ and /eɪ/, were analyzed using Euclidean distance in the F1–F2 vowel space. The results showed that the distance between the onset /a/ and the offset /ɪ/ was larger when speech was produced without the mask and when speaking clearly. These modifications were larger for L1 talker compared to the L2 talker. The Euclidian distance for /eɪ/was only affected by speaking style. The results suggest that talkers produced hyperarticulated diphthongs characterized by larger formant movements in clear speech. The presence of a mask limited jaw movement for the diphthong containing the low vowel. Additionally, the L1 talker made larger articulatory modifications in response to the presence of a mask and listener-oriented clear speech compared to the talker with less extensive experience with the target language. These production patterns may be related to lower word recognition in noise for masked, conversational, and L2 speech found in previous work.

Coarticulation is reduced in clear speech produced with protective face masks

Talkers dynamically modify their coarticulatory patterns when producing listener-oriented hyperarticulated clear speeches. This study examined how the use of protective face masks interacts with the production of intelligibility-enhancing clear speech to impact coarticulation. A native and a non-native speaker of English read sentences in a clear and conversational speaking style with and without a surgical mask. Coarticulation between word-internal adjacent segments was analyzed with a whole-spectrum analysis including spectral distance and segment overlap duration. Both speakers coarticulated less in clear than in conversational speaking style as indicated by the larger spectral distance and shorter overlap duration between adjacent segments. Coarticulation was further reduced when clear speech was produced with a mask by the native speaker but not by the non-native speaker. The findings showed that producing hyperarticulated intelligibility-enhancing clear speech also involves reducing coarticulatory overlap across adjacent segments. Coarticulatory resistance was adaptively reinforced in the presence of the additional communicative barrier, face mask, particularly for the speaker with extensive experience with the target language. Such cumulative reduction of coarticulation may in part underlie the larger perception-in-noise benefit for clear speech produced with a mask for the native compared to the non-native talker.

Clear speech promotes speaking rate normalization

When speaking in noisy conditions or to a hearing-impaired listener, talkers often use clear speech which is slower, louder, and hyperarticulated relative to conversational speech. In other research, changes in speaking rate are known to affect speech perception (called temporal contrast effects, or speaking rate normalization). For example, when a sentence is spoken quickly, the voice onset time (VOT) in the next word sounds longer by comparison (e.g., more like the /t/ of “tier”); a sentence spoken slowly makes the VOT in the next word sound shorter (e.g., like the /d/ of “deer”). Typically, one sentence is manipulated to produce fast and slow versions. We tested whether naturally produced clear and conversational speaking styles would also produce these temporal contrast effects. On each trial, listeners heard either a clear (slow) sentence or a conversational (fast) sentence followed by a target word to be categorized as “deer” or “tier.” Temporal contrast effects were observed both for conversational relative to clear speech and for conversational relative to a slowed version of the conversational speech. Changing speaking styles aids speech intelligibility but may produce other consequences such as contrast effects that alter sound/word recognition.

Reading aloud in clear speech reduces sentence recognition memory and recall for native and non-native talkers.

Speaking style variation plays a role in how listeners remember speech. Compared to conversational sentences, clearly spoken sentences were better recalled and identified as previously heard by native and non-native listeners. The present study investigated whether speaking style variation also plays a role in how talkers remember speech that they produce. Although distinctive forms of production (e.g., singing, speaking loudly) can enhance memory, the cognitive and articulatory efforts required to plan and produce listener-oriented hyper-articulated clear speech could detrimentally affect encoding and subsequent retrieval. Native and non-native English talkers' memories for sentences that they read aloud in clear and conversational speaking styles were assessed through a sentence recognition memory task (experiment 1; N = 90) and a recall task (experiment 2; N = 75). The results showed enhanced recognition memory and recall for sentences read aloud conversationally rather than clearly for both talker groups. In line with the "effortfulness" hypothesis, producing clear speech may increase the processing load diverting resources from memory encoding. Implications for the relationship between speech perception and production are discussed.

The Clear-Speech Benefit for School-Age Children: Speech-in-Noise and Speech-in-Speech Recognition.

Purpose Talkers often modify their speech when communicating with individuals who struggle to understand speech, such as listeners with hearing loss. This study evaluated the benefit of clear speech in school-age children and adults with normal hearing for speech-in-noise and speech-in-speech recognition. Method Masked sentence recognition thresholds were estimated for school-age children and adults using an adaptive procedure. In Experiment 1, the target and masker were summed and presented over a loudspeaker located directly in front of the listener. The masker was either speech-shaped noise or two-talker speech, and target sentences were produced using a clear or conversational speaking style. In Experiment 2, stimuli were presented over headphones. The two-talker speech masker was diotic (M0). Clear and conversational target sentences were presented either in-phase (T0) or out-of-phase (Tπ) between the two ears. The M0Tπ condition introduces a segregation cue that was expected to improve performance. Results For speech presented over a single loudspeaker (Experiment 1), the clear-speech benefit was independent of age for the noise masker, but it increased with age for the two-talker masker. Similar age effects for the two-talker speech masker were seen under headphones with diotic presentation (M0T0), but comparable clear-speech benefit as a function of age was observed with a binaural cue to facilitate segregation (M0Tπ). Conclusions Consistent with prior research, children showed a robust clear-speech benefit for speech-in-noise recognition. Immaturity in the ability to segregate target from masker speech may limit young children's ability to benefit from clear-speech modifications for speech-in-speech recognition under some conditions. When provided with a cue that facilitates segregation, children as young as 4-7 years of age derived a clear-speech benefit in a two-talker masker that was similar to the benefit experienced by adults.

Talker differences in clear and conversational speech: Perceived emotional valence

Clear speech, a speaking style talkers adopt when talking to individuals with hearing loss, is often described as sounding “angry” (e.g., “I try to speak clearly, but when I do, he thinks I'm mad at him”). Previous research has indicated that listeners do indeed rate clear speech as “angry” or “disgusted” more often than conversational speech. However, this same research found that some talkers can increase the clarity of their speech without sounding angry. The present study expands the previous work by testing perceived emotion for 41 talkers in order to examine talker differences in the perceived emotional valence of clear and conversational speech. Perceptual ratings from 25 young adults with normal hearing will be gathered using an online experiment interface. Stimuli consist of 14 emotionally neutral sentences in both clear and conversational speaking styles from 41 talkers. Listeners will be asked to “judge the emotion you think you heard in that sentence” using a six-alternative, forced-choice paradigm for thefollowing emotional categories: “anger,” “fear,” “disgust,” “sadness,” “happiness,” and “neutral.” It is anticipated that the current study will increase our understanding of how perceived emotional valence in clear and conversational speaking styles varies among talkers.

Word-final T-deletion in Southern German

The recordings of 20 native German speakers were analysed to identify the strongest factors affecting the rates of word-final t-deletion. Deletion rates were shown to be higher for /t/s in semiweak verbs, when preceded or followed by sibilants, and in a conversational speaking style. In addition to this, frequent words showed higher deletion rates. English and German t-deletion were comparable to some extent in this study. However, deletion rates by morphological complexity showed differences between the languages: monomorphemic words, in particular, had lower deletion rates in German than in English. It was also shown that reading aloud reduces the deletion rates significantly which is in line with previous research.

Do speakers maintain clear speech throughout natural conversation?

Previous studies have shown that speakers tend to speak more clearly when they speak with a listener in adverse listening conditions in naturalistic settings. Most previous studies assumed that when speakers speak more clearly, they maintain the style throughout the conversation. However, a clear speech style deviates from speakers’ conversational speaking style, and extra effort is required in order to produce clear speech. Therefore, it is possible that speakers begin a conversation in a clear speaking style and revert back to a conversational style at some point during the conversation. This study aims to examine whether native English speakers use and maintain a clear speaking style while participating in a conversation with non-native speakers in naturalistic settings. Native English listeners will respond to sentences from naturalistic conversations. The sentences are from either a conversation between two native English speakers or a conversation between a native English speaker and a non-native English speaker. Intelligibility (typing what the listeners heard) and comprehensibility (rating how easy it is to understand the sentences) are measured to examine the usage of clear speech throughout the conversations. The results of this study will provide a better understanding of the circumstances under which speakers use clear speech.Previous studies have shown that speakers tend to speak more clearly when they speak with a listener in adverse listening conditions in naturalistic settings. Most previous studies assumed that when speakers speak more clearly, they maintain the style throughout the conversation. However, a clear speech style deviates from speakers’ conversational speaking style, and extra effort is required in order to produce clear speech. Therefore, it is possible that speakers begin a conversation in a clear speaking style and revert back to a conversational style at some point during the conversation. This study aims to examine whether native English speakers use and maintain a clear speaking style while participating in a conversation with non-native speakers in naturalistic settings. Native English listeners will respond to sentences from naturalistic conversations. The sentences are from either a conversation between two native English speakers or a conversation between a native English speaker and a non-native En...

Acoustic-phonetic consequences of clear speech in Cantonese vowels and lexical tones

In adverse communicative contexts, speakers will modify their speech style to enhance intelligibility, which can involve a range of acoustic and articulatory adjustments. Prior work has focused on the influence of speaking style on spectral (e.g., vowel space expansion) and temporal distinctions (e.g., voice-onset time). Thus, this study investigated the extent to which such adjustments extend to the pitch dimension, namely lexical tone. Ten native Cantonese speakers were recorded producing five monosyllables with six Cantonese tones embedded in sentence contexts in both clear and conversational speaking styles. Acoustic analyses revealed that speakers produced two distinct styles, with longer vowel durations and greater vowel dispersion in clear relative to conversational speech. However, with regards to lexical tone production, the results indicated only a marginal re-adjustment of tones within the tone space as a result of clear speech. Specific contour tones, such as the low- and high-rising tones, saw increases in their F0 offset. However, mean F0, F0 range and tonal dispersion did not differ substantially between styles. These findings indicate that speakers maintain tonal stability across speaking styles, suggesting that, relative to segments, lexical tones may not be prioritized as phonological contrasts vital to enhance the overall intelligibility of the utterance.

Effect of proficiency on perception of English nasal stops by native speakers of Brazilian Portuguese

Brazilian Portuguese (BP) does not contrast /m/ and /n/ in coda position, leading to perceptual difficulties with minimal pairs like lime/line in L2 English. This study examined the effect of acoustic-phonetic and semantic enhancements on word recognition in noise for native English (n = 26) and L1 BP/L2 English (n = 21) listeners. They heard 120 high- and low-predictability sentences produced in clear and conversational speaking styles and mixed with speech-shaped noise. They typed the last word ending in either /m/ or /n/. The L2 listeners were separated into high- and low-proficiency (HP and LP) groups based on the perceived accentedness ratings of speech samples they provided. The results showed that native listeners benefited more from semantic than acoustic-phonetic enhancements alone, and the accuracy was highest when both enhancements were combined. The HP L2 listeners recognized more target words overall than the LP group, and their benefit from semantic over acoustic-phonetic enhancements mirrored that of the native listeners. In contrast, LP listeners could not utilize the higher-level semantic information to the same degree in the absence of signal clarity. These findings show that L2 listeners benefited from acoustic-phonetic, semantic, and combined enhancements when recognizing /m/-/n/ coda pairs, but this effect was mediated by proficiency.

Semantic context modulates intelligibility advantage of clear speech in temporally compressed sentences

Sentences produced in a clear speech speaking style are typically more intelligible than sentences produced in a conversational style. However, factors that lead to the intelligibility difference are not fully understood. This study investigated the role of semantic context in intelligibility of temporally compressed clear and conversational speech. Grammatically correct English sentences, which were either semantically meaningful or semantically anomalous, were presented to 70 normal-hearing listeners in clear or conversational speaking style. The sentences were temporally compressed to produce seven different syllabic rates (6-12), matched across clear and conversational sentences to control for the original durational differences between clear and conversational speech. Results for semantically anomalous sentences indicate no differences in the rate of intelligibility decline across compression rates for clear and conversational speech. In contrast, for semantically meaningful sentences, the rate of intelligibility decline was smaller for the semantically meaningful clear-speech sentences over semantically anomalous conversational-speech sentences. At the highest compression rate of 12 syllables per second, intelligibility of semantically meaningful clear-speech sentences (~68% correct) was similar to that of semantically anomalous sentences at 9 syllables per second. These findings suggest that semantic context can modulate the effect of speaking style on the intelligibility of temporally compressed sentences.

Variation in normal hearing thresholds predicts word recognition in noise

Analyses of speech identification data collected from young, normal-hearing adults often do not include measures of listeners’ hearing. Sub-clinical variation in hearing, it is assumed, will not significantly affect the identification of speech signals presented well above threshold levels. This study tested that assumption in the context of an experiment that investigated young adults’ recognition of words with few versus many phonological neighbors produced in clear and conversational speaking styles. Words were presented in speech-shaped noise at a + 5 dB SNR. A mixed effects regression analysis was performed to account for the contributions of lexical difficulty, speaking style, order of word presentation, and listeners’ hearing (as measured by pure-tone average (PTA)). This analysis showed, as expected, that correct identification was predicted by word type (easy > hard) and speaking style (clear > conversational). It also showed that identification improved over the course of the experiment. Finally...

Multi-stream LSTM-HMM decoding and histogram equalization for noise robust keyword spotting

Highly spontaneous, conversational, and potentially emotional and noisy speech is known to be a challenge for today's automatic speech recognition (ASR) systems, which highlights the need for advanced algorithms that improve speech features and models. Histogram Equalization is an efficient method to reduce the mismatch between clean and noisy conditions by normalizing all moments of the probability distribution of the feature vector components. In this article, we propose to combine histogram equalization and multi-condition training for robust keyword detection in noisy speech. To better cope with conversational speaking styles, we show how contextual information can be effectively exploited in a multi-stream ASR framework that dynamically models context-sensitive phoneme estimates generated by a long short-term memory neural network. The proposed techniques are evaluated on the SEMAINE database-a corpus containing emotionally colored conversations with a cognitive system for "Sensitive Artificial Listening".

Effect of language, speaking style and talker on a spectral rhythm measure.

Recently, two spectro‐temporal approaches have been proposed for characterizing cross‐language and disordered speech rhythm: power spectrum [Tilsen and Johnson (2008)] and speech envelope modulation spectrum (EMS) [Liss et al. (2010)]. These measures quantify the rhythmicity of speech within specified frequency bands. In the present study, the EMS measure was employed to assess the effect of language, speaking style, and talker on rhythm. In order to examine the rhythmic characteristics of conversational and speaking styles in English and Croatian, utterances produced by two female speakers of each language in clear and conversational speaking styles were analyzed [Smiljanic and Bradlow (2005)]. The spectral centroid, a weighted mean of a spectrum, was used to measure the center of mass of the EMS extracted for each utterance. Results from nested factorial analyzes revealed that the spectro‐temporal characteristics differed for English and Croatian, for the conversational and clear speaking sentences, and for individual talkers. These results suggest that the low rate amplitude modulations of a signal can be used to distinguish rhythmic properties across languages, speaking styles, and talkers. Furthermore, these automated spectral measures provide a new means for characterizing rhythm that bypasses laborious manual measurements and phonologically dependent segmentation criteria.

Applied principles of clear and Lombard speech for automated intelligibility enhancement in noisy environments

Previous studies have documented phenomena involving the modification of human speech in special communication circumstances. Whether speaking to a hearing-impaired person (clear speech) or in a noisy environment (Lombard speech), speakers tend to make similar modifications to their normal, conversational speaking style in order to increase the understanding of their message by the listener. One strategy characteristic of the above speech types is to increase consonant power relative to the signal power of adjacent vowels and is referred to as consonant–vowel (CV) ratio boosting. An automated method of speech enhancement using CV ratio boosting is called energy redistribution voiced/unvoiced (ERVU). To characterize the performance of ERVU, 25 listeners responded to 500 words in a two-word, forced-choice experiment in the presence of energetic masking noise. The test material was a vocabulary of confusable monosyllabic words spoken by 8 male and 8 female speakers, and the conditions tested were a control (unmodified speech), ERVU, and a high-pass filter (HPF). Both ERVU and the HPF significantly increased recognition accuracy compared to the control. Nine of the 16 speakers were significantly more intelligible when ERVU or the HPF was used, compared to the control, while no speaker was less intelligible. The results show that ERVU successfully increased intelligibility of speech using a simple automated segmentation algorithm, applicable to a wide variety of communication systems such as cell phones and public address systems.

Landmark analysis of clear and conversational speaking styles

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.

Variability in the clear speech intelligibility advantage

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.

Tone clarity in mixed pitch/phonation type tones

Lexical tone identity is often determined by a complex of acoustic cues. In Green Mong, a Hmong-Mien language of Southeast Asia, a small subset of tones is characterized by phonation type in addition to pitch height, pitch contour, and duration, which characterize the remaining tones of the language. In tones that incorporate multiple cues to tonal identity, what makes a tone clear, or easy to recognize? This study examines acoustic and perceptual data to address this question. Six native speakers of Green Mong were asked to produce 132 phonological CV words in sentence context, using a conversational speaking style. Seventeen native speakers of the language were then asked to categorize three tones which have similar falling contours, but are differentiated by phonation type (breathy, creaky, and modal). Tokens that were correctly identified by 100% of the listeners were compared with tokens that were relatively poorly identified. Data indicate that the breathy- and creaky-voiced tones are less susceptible to identification errors than the modal-voiced tone. However, the clearest tokens of the three tones are also differentiated by details of pitch contour shape, and by duration. Similarities and differences between acoustic cue values for the best and worst tokens will be discussed.