Search

Backchannels (BCs) are signals produced by conversation partners to support the ongoing turn of the interlocutor. Audible backchannels can have a lexical (yes) or non-lexical form (mhm), consisting of one unit (mhm) or multiple units (mhm yes). In this paper, we analyse the form and function of Multi-Unit Backchannels (MUBs) in Italian conversational speech. Data are drawn from the DIA (Dialogic ItAlian) corpus, a collection of spontaneous and informal conversations between dyads of Italian speakers who know each other well. In this corpus, MUBs represent over 29% of total backchanneling signals. While most single-unit backchannels fulfil a single function, with Continuer being the most frequent, the majority of MUBs fulfil multiple functions simultaneously, usually involving Agreement. Although backchannels are regarded as very short utterances, a large proportion of MUBs are, in fact, not short in duration. Interestingly, in our corpus MUBs are produced more frequently by speakers who do not take a prominent role in the interaction. • Multi-Unit Backchannels (MUBs) are prevalent in Italian informal conversational speech. • MUBs constitute 29% of all backchannels in our corpus. • MUBs fulfil multiple functions simultaneously, usually involving Agreement. • Over 68% of MUBs have a duration longer than 500 ms [ABSTRACT FROM AUTHOR]

Speech is a crucial aspect of human-to-human interactions and plays a fundamental role in the advancement of human–computer interaction (HCI) systems. Developing an accurate speech emotion recognition (SER) system for human conversations poses a critical yet challenging task. Existing state-of-the-art (SOTA) research in SER primarily focuses on modeling vocal information within individual conversational speech utterances, overlooking the significance of incorporating transactional information from the interaction context. In this paper, we present a novel Contextualized Convolutional Transformer-GRU Network (CCTG-Net) for recognizing speech emotions using Mel-spectrogram features, effectively integrating contextual information for emotion recognition. Our experiments are conducted on the widely-used emotional benchmark dataset, IEMOCAP. Compared to SOTA methods in four-class emotion recognition, our proposed model achieves a weighted accuracy of 88.4% and an unweighted accuracy (UA) of 89.1%. This marks a substantial 3.0% enhancement in UA while maintaining an optimal balance between performance and complexity. [ABSTRACT FROM AUTHOR]

Highly performing speech recognition is important for more fluent human–machine interaction (e.g., dialogue systems). Modern ASR architectures achieve human-level recognition performance on read speech but still perform sub-par on conversational speech, which arguably is or, at least, will be instrumental for human–machine interaction. Understanding the factors behind this shortcoming of modern ASR systems may suggest directions for improving them. In this work, we compare the performances of HMM- vs. transformer-based ASR architectures on a corpus of Austrian German conversational speech. Specifically, we investigate how strongly utterance length, prosody, pronunciation, and utterance complexity as measured by perplexity affect different ASR architectures. Among other findings, we observe that single-word utterances – which are characteristic of conversational speech and constitute roughly 30% of the corpus – are recognized more accurately if their F0 contour is flat; for longer utterances, the effects of the F0 contour tend to be weaker. We further find that zero-shot systems require longer utterance lengths and are less robust to pronunciation variation, which indicates that pronunciation lexicons and fine-tuning on the respective corpus are essential ingredients for the successful recognition of conversational speech. [ABSTRACT FROM AUTHOR]

This paper presents a corpus-based perspective on the phonetic fusion of disyllabic words in a Chinese conversational speech corpus. Four categorical types that reflect the phonological features of reduction degrees are automatically derived from gradient, acoustic properties. A transcription experiment is conducted with the most common disyllabic words. Both automatic derivation by acoustic signals and human transcription by perceptual judgment refer to the same sound inventory. We have shown that the complete form of fusion occurring in conversation need not be legitimate syllables and it appears consistently in the form of syllable merger that represents a group of phonetic variants. [ABSTRACT FROM AUTHOR]

Previous research has shown that differences in the marking of prosodic boundaries exist not only between varieties of the same language, but also between speaking styles of the same variety. Our aim is to gain insights into how these two factors interact in the case of German and Austrian German read and conversational speech. Analysing four acoustic cues, representing durational and fundamental frequency (f0) measures, we observed that pause duration was the strongest cue to prosodic boundaries and that f0 reset was the weakest, in both varieties and across speaking styles. In read speech however, we noticed differences between the two varieties in the weighting given to the four cues. We then employed the four investigated acoustic cues for the annotation of the less-resourced variety, by examining ways of exploiting resources from the more well-resourced variety. Testing three automatic approaches for boundary detection, we obtained an overall high performance: 82.4% and 76.6% F-score for read and conversational speech, respectively. Moreover, our results showed that training on more data from the well-resourced variety did not outperform the same system trained on less data from the target variety. • Prosodic boundaries in German and Austrian German, read and conversational speech. • Pause is the strongest cue, f0 reset is the weakest. • Cues weighting is more similar in conversational speech between the two varieties. • Automatic detection systems using the cues performed well on both speaking styles. • Training in variety-mismatched condition was not better than matched condition. [ABSTRACT FROM AUTHOR]

Unusual speech prosody has long been recognized as a characteristic feature of the speech of individuals diagnosed with Autism Spectrum Disorders (ASD). However, research to determine the exact nature of this difference in speech prosody is still ongoing. Many individuals with verbal autism perform well on tasks testing speech prosody. Nonetheless, their expressive prosody is judged to be unusual by others. We propose that one aspect of this perceived difference in speech prosody in individuals with ASD may be due to a deficit in the ability to entrain—or become more similar—to their conversation partners in prosodic features over the course of a conversation. In order to investigate this hypothesis, 24 children and teens between the ages of 9 and 15 years participated in our study. Twelve of the participants had previously been diagnosed with ASD and the other 12 participants were matched to the ASD participants in age, gender, and non-verbal IQ scores. All participants completed a goal-directed conversation task, which was subsequently analyzed acoustically. Our results suggest (1) that youth diagnosed with ASD entrain less to their conversation partners compared to their neurotypical peers—in fact, children and teens diagnosed with ASD tend to dis-entrain from their conversation partners while their neurotypical peers tend to converge to their conversation partners’ prosodic features. (2) Although age interacts differently with prosodic entrainment in youth with and without ASD, this difference is attributable to the entrainment behavior of the conversation partners rather than to those with ASD. (3) Better language skill is negatively correlated with prosodic entrainment for both youth with and without ASD. The observed differences in prosodic entrainment in children and teens with ASD may not only contribute to the perceived unusual prosody in youth with ASD but are also likely to be indicative of their difficulties in social communication, which constitutes a core challenge for individuals with ASD.

A previous investigation demonstrated differences between younger adult normal-hearing listeners and older adult hearing-impaired listeners in the perceived emotion of clear and conversational speech. Specifically, clear speech sounded angry more often than conversational speech for both groups, but the effect was smaller for the older listeners. These listener groups differed by two confounding factors, age (younger vs. older adults) and hearing status (normal vs. impaired). The objective of the present study was to evaluate the contributions of aging and hearing loss to the reduced perception of anger in older adults with hearing loss. We investigated perceived anger in clear and conversational speech in younger adults with and without a simulated age-related hearing loss, and in older adults with normal hearing. Younger adults with simulated hearing loss performed similarly to normal-hearing peers, while normal-hearing older adults performed similarly to hearing-impaired peers, suggesting that aging was the primary contributor to the decreased anger perception seen in previous work. These findings confirm reduced anger perception for older adults compared to younger adults, though the significant speaking style effect—regardless of age and hearing status—highlights the need to identify methods of producing clear speech that is emotionally neutral or positive.

Objectives: One of the main prosodic components of language is the rate. Speaking and articulation rates are two different measurements that reflect various aspects of each other. This study mainly aimed to present preliminary normative data related to speaking rate, reading rate, articulation rate and articulation rate in reading for Turkish-speaking adults and also aimed to compare these four measurements in terms of gender. Methods: The present study included 84 university students (42 males and 42 females) aged between 19-24 years old whose native language was Turkish. Power analysis was calculated based on the articulation rate. Speech and articulation rates were measured by taking 400-syllable conversational speech samples from each participant. A text was used from the Adult II section of the Turkish version of the Stanford Binet Intelligence Test to measure their reading rates. The speaking and reading samples obtained were recorded with a voice recorder and analyzed with the PRAAT software. Results: There was a statistically significant difference both between their speaking and reading rates, and between articulation rate and articulation rate in reading. Male participants had significantly higher articulation rates in speaking and reading than females. Conclusion: The present study provided evidence that the speech rate is low and articulation rate is high in Turkish. It concluded that the significant difference between articulation rates by gender in many languages is also valid for Turkish. [ABSTRACT FROM AUTHOR]

Unusual speech prosody has long been recognized as a characteristic feature of the speech of individuals diagnosed with Autism Spectrum Disorders (ASD). However, research to determine the exact nature of this difference in speech prosody is still ongoing. Many individuals with verbal autism perform well on tasks testing speech prosody. Nonetheless, their expressive prosody is judged to be unusual by others. We propose that one aspect of this perceived difference in speech prosody in individuals with ASD may be due to a deficit in the ability to entrain—or become more similar—to their conversation partners in prosodic features over the course of a conversation. In order to investigate this hypothesis, 24 children and teens between the ages of 9 and 15 years participated in our study. Twelve of the participants had previously been diagnosed with ASD and the other 12 participants were matched to the ASD participants in age, gender, and non-verbal IQ scores. All participants completed a goal-directed conversation task, which was subsequently analyzed acoustically. Our results suggest (1) that youth diagnosed with ASD entrain less to their conversation partners compared to their neurotypical peers—in fact, children and teens diagnosed with ASD tend to dis-entrain from their conversation partners while their neurotypical peers tend to converge to their conversation partners' prosodic features. (2) Although age interacts differently with prosodic entrainment in youth with and without ASD, this difference is attributable to the entrainment behavior of the conversation partners rather than to those with ASD. (3) Better language skill is negatively correlated with prosodic entrainment for both youth with and without ASD. The observed differences in prosodic entrainment in children and teens with ASD may not only contribute to the perceived unusual prosody in youth with ASD but are also likely to be indicative of their difficulties in social communication, which constitutes a core challenge for individuals with ASD. [ABSTRACT FROM AUTHOR]

Stress causes a speaker's voice characteristics to be changed. Emotional stress alters a person's speech pattern such that it is distributed non-normally along the temporal dimension. Thus, the methods for identifying the gender of a non-stressed speaker were no longer effective in recognizing the gender of a speaker in stressful conditions. To address this issue, a new gender identification framework is proposed. We leveraged i-vector for capturing gender information on each speech segment. Then the long short-term memory dynamically handled all speech temporal context features and learned the long-term dependency from the input. We evaluated the effectiveness, in terms of accuracy and the number of iterations to saturate, of the proposed method by comparing it with the baseline methods in their respective abilities to identify the speaker's gender from conversations with different durations. By learning the gender information encoded in long-term dependencies, our proposed method outperforms the baseline methods and is able to correctly identify the speaker's gender in all conversation types. [ABSTRACT FROM AUTHOR]

This study investigates the psycholinguistic process underlying triggered codeswitching – codeswitching facilitated by the occurrence of cognates – within the context of conversational dynamics. It confirms that, in natural bilingual speech, lexical selection of cognates can facilitate codeswitching by enhancing the activation of the non-selected language. Analyses of a large-scale corpus of Welsh–English conversational speech showed that 1) producing cognates facilitated codeswitching, 2) speakers who generally produced more cognates generally codeswitched more, even in clauses that did not contain cognates, 3) larger numbers of cognates in a clause increased the likelihood of codeswitching, 4) codeswitching temporarily remained facilitated after the production of cognates, and 5) hearing rather than producing cognates did not facilitate codeswitching. The findings confirm the validity of the proposed cognitive account of triggered codeswitching, and clarify the relation between the lexical activation of cognates and consecutive language choice, in accord with current insights in lexical processing. [ABSTRACT FROM AUTHOR]

The relative contributions of static and dynamic formant representations to speaker- specificity were investigated in conversational speech and in two vowels varying in inherent spectral change. Using polynomial fits, the contribution of dynamic formant coefficients to speaker-specificity relative to that of the formant intercept was investigated in the diphthongal vowel [ei] taken from English and Dutch conversational speech. The [ei] tokens were sampled from various linguistic contexts and analysed in an LR approach. Results show that formant dynamics contain speaker-specific information in conversational speech even though the high contextual variation seems to reduce its effect relative to that reported by earlier work. Vowels differ in inherent dynamicity and therefore, the added value of dynamic formant information to speaker-specificity was also compared between vowels differing in inherent spectral change. Using Dutch data, the contribution of formant dynamics to speaker-specificity was compared between [ei] and [aː] tokens produced by the same speakers. Formant dynamics in conversational speech only contributed to speaker-specificity in the diphthong [ei], not in the monophthong [aː]. [ABSTRACT FROM AUTHOR]

In spoken language, reductions of word forms occur regularly and need to be accommodated by the listener. Intriguingly, this accommodation is usually achieved without any apparent effort. The neural bases of this cognitive skill are not yet fully understood. We here presented participants with reduced words that were either preceded by a related or an unrelated visual prime and compared electric brain responses to reduced words with those to their full counterparts. In time-domain, we found a positivity between 400 and 600 ms differing between reduced and full forms. A later positivity distinguished primed and unprimed words and was modulated by reduction. In frequency-domain, alpha suppression was stronger for reduced than for full words. The time- and frequency-domain reduction effects converge towards the view that reduced words draw on attention and memory mechanisms. Our data demonstrate the importance of interactive processing of bottom-up and top-down information for the comprehension of reduced words. [ABSTRACT FROM AUTHOR]

Highlights • Stuttering occurrence in spontaneous speech was related to phonological network characteristics. • Adults stuttered on words with fewer phonological neighbors (sparse neighborhood density). • Phonological encoding difficulty appears to play a contributing role in the occurrence of stuttering in spontaneous speech in adults who stutter. Abstract Purpose The study examined the relationship between word-form similarity network (phonological neighborhood) and stuttering occurrence in spontaneous speech in adults. The study asked whether neighborhood characteristics, including the number of neighbors (neighborhood density) and the average word frequency among neighbors (neighborhood frequency), differentiate stuttered from fluent words within spontaneous speech samples, and more specifically, whether neighborhood characteristics facilitate speech fluency in adults who stutter. Method Spontaneous speech samples were collected from 14 adults who stutter, including 7 with mild stuttering and 7 with severe stuttering. Each stuttered word was matched with a fluently produced word within the sample, controlling for multiple lexical factors. Neighborhood density and neighborhood frequency were compared between the stuttered and matched fluent words. Results Adults stuttered on words with lower neighborhood density, compared to fluently produced words, with similar patterns in the two severity subgroups. There appeared to be a marginal difference between single-syllable whole-word repetition and the other stuttering typologies. Conclusion Neighborhood density facilitates fluent word production in spontaneous speech in adults who stutter, regardless of stuttering severity. This finding adds to the evidence supporting that phonological encoding plays a role in stuttering in naturalistic, spontaneous speech in adults. [ABSTRACT FROM AUTHOR]

This study investigates the plasticity of phonological boundaries during discourse in a lingua franca. We tracked the production of 34 Spanish learners of English conversing with two Dutch confederates in English across two speech styles, focusing on incremental changes in two key English vowel contrasts with differential effects of cross-linguistic influence (/i/-/I/ and /ε/-/æ/). Results indicate that Spaniards align with Dutch confederates, quickly merging /ε/ and /æ/ and gradually separating their merged /i/-/I/ category, rather than adopting native-like English production. We found greater merger in informal speech overall. We also found an interaction with time for the /i/-/I/ contrast, indicating that the merged /i/ and /I/ categories gradually separate in informal speech; this effect was not found for /ε/-/æ/. Finally, proficiency modulates alignment: the most proficient speakers separate /i/-/I/ and merge /ε/-/æ/ more than other speakers. We interpret phonetic alignment as a complex, dynamic phenomenon influenced by proficiency in discourse language and speaking style, and whose effects may unfold rapidly or gradually depending on the phonological category investigated. [ABSTRACT FROM AUTHOR]

Copyright of Itinerarios (1507-7241) is the property of University of Warsaw and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)