Fabrice Hirsch

Stuttering is a neurodevelopmental disorder marked by impaired speech production and substantial interindividual variability in symptom severity. Although neuroimaging studies have identified abnormalities in the basal ganglia-thalamocortical circuitry, the specific white matter pathways that underpin dysfluent speech production remain poorly understood. Here, we focus on the frontal aslant tract (FAT), a dorsoventral intralobar tract connecting the superior and inferior frontal cortices and proposed to support both cognitive and motor functions. Using diffusion MRI, we first compared the quantitative anisotropy of the FAT and relevant control tracts in individuals who stutter (n = 30, mean age = 35.67, SD = 13.03) and fluent speakers (n = 22, mean age = 35.64, SD = 10.87) and observed no significant group-level differences in either hemisphere. To delineate the role of FAT subbundles in speech dysfluency, we then segmented the tract at its dorsal and ventral cortical terminations and assessed their relationships with stuttering severity (Stuttering Severity Instrument-Fourth Edition). We identified left FAT subbundles connecting the posteroventral frontal structures, particularly the rostroventral precentral gyrus and, to a lesser extent, the pars opercularis to a dorsal prefrontal region whose integrity predicts stuttering severity. The left-lateralized nature of these associations suggests that stuttering results, in part, from abnormal connectivity between prefrontal systems governing volitional control and the motor circuits engaged in speech production. Together, these findings refine current neurobiological models of stuttering by implicating frontofrontal white matter pathways underlying speech dysfluency and highlight a potential anatomical target for individualized therapeutic interventions.

Background. Speech analysis can detect subtle cognitive impairment, particularly because prosody contains fine elements, such as pauses, which might be behavioral markers of NeuroCognitive Disorders. However, the absence of simple, detailed methods compromises the feasibility of such an analysis in clinics. Mild Neurocognitive Disorders (mNCD) refer to a cognitive change reported by an individual whose autonomy is preserved. This terminology covers many underlying pathophysiologies, such as Cancer-Related Cognitive Impairment (CRCI). CRCI is an mNCD characterized by a memory complaint (e.g., "I forget what I’m told") and a language complaint (e.g., "I search for my words") for which the frequency and intensity exceed the scores of speech-language tests or neuropsychological tests. The lack of sensitive tools to detect this subtle impairment often leads to its underdiagnosis. New methods such as speech analysis are needed to assess CRCI. Objectives. The study aims are (i) to propose a method for analyzing pauses which is practical in clinical context, (ii) to identify participants with CRCI using this method. Methods. Thirteen participants post breast cancer and thirteen healthy controls were included in the study. The participants were instructed to tell a picture-based story. Their narratives were recorded, then automatically transcribed with Whisper and analyzed using SPPAS and Praat software. Silent pauses, filled pauses (e.g., "uh"), and sustained vowels (e.g., "a smaall boy") were annotated, then examined on JASP for a statistical analysis. Results. The duration of silent pauses in post-cancer participants was significantly longer than that of the controls. However, the duration of filled pauses and sustained vowels did not show a significant difference between the two groups. Similarly, the rate of pauses (i.e., number of occurrences/total speech duration) did not show a significant difference between the two groups. Conclusions. This study provides a well-described method for pause analysis designed for the purpose of clinical context. The results suggest that the duration of silent pauses is a good marker for differentiating post-cancer participants from controls.

Historically, French has been a significant international language, with its influence apparent in global vocabulary borrowings. However, by the latter half of the 20th century, geopolitical shifts saw English assuming the role of the global lingua franca, thereby influencing various languages, including French. The fashion industry vividly displays this shift in French vocabulary. In earlier centuries, French dominated fashion terminology, influencing even Czech with words like “kravata” (cravate) and “bižutérie” (bijouterie). Presently, as English gains dominance in global communication, there is a surge in Anglicisms in French. Designers now frequently use English-based neologisms, believing them to heighten product appeal. France’s advanced linguistic policy, led by the Commission d’enrichissement de la langue française, works to counter this trend by providing native French equivalents, especially in economic, legal, and scientific fields. Interestingly, while Quebec’s linguistic body works in tandem with France’s, there are differences, like the term for “hashtag”: France recommends “mot-dièse” whereas Quebec suggests “mot-clic”. This paper aims to contrast the use of fashion-related Anglicisms and their official recommendations across varied linguistic tools.

Introduction: Mild Neurocognitive Disorders (mNCD) are characterized by a subtle decline in cognitive performance. Discourse analysis is a research method that investigates the verbal production of patients and has been used to screen for subtle cognitive impairment. Studies using discourse analysis are, however, rarely replicated. One reason for this is likely to be the lack of consistency across studies regarding the selection and definition of relevant linguistic features. The goal of this review is to identify reliable linguistic markers for the detection of subtle cognitive impairment in connected speech, and generate hypotheses associating these markers with underlying cognitive functions in a given discourse context. Methods and analysis: Following the PRISMA guidelines, we will conduct a systematic review including five databases: PubMed, ScienceDirect, Embase, Web of Science, Google Scholar. We will include articles targeting a population of patients with subtle cognitive impairment, such as Mild Cognitive Impairment. Eligible articles will include at least one connected speech task and an analysis of linguistic features in this task. The data will be classified according to study population, connected speech task, analysed linguistic features, and methods used to compare cases and controls. PROSPERO registration number: CRD42023394729.

The adoption of advanced deep learning architectures in stuttering detection (SD) tasks is challenging due to the limited size of the available datasets. To this end, this work introduces the application of speech embeddings extracted from pre-trained deep learning models trained on large audio datasets for different tasks. In particular, we explore audio representations obtained using emphasized channel attention, propagation, and aggregation time delay neural network (ECAPA-TDNN) and Wav2Vec2.0 models trained on VoxCeleb and LibriSpeech datasets respectively. After extracting the embeddings, we benchmark with several traditional classifiers, such as the K-nearest neighbour (KNN), Gaussian naive Bayes, and neural network, for the SD tasks. In comparison to the standard SD systems trained only on the limited SEP-28k dataset, we obtain a relative improvement of 12.08%, 28.71%, 37.9% in terms of unweighted average recall (UAR) over the baselines. Finally, we have shown that combining two embeddings and concatenating multiple layers of Wav2Vec2.0 can further improve the UAR by up to 2.60% and 6.32% respectively.

Stuttering is a neuro-developmental speech impairment characterized by uncontrolled utterances (interjections) and core behaviors (blocks, repetitions, and prolongations), and is caused by the failure of speech sensorimotors. Due to its complex nature, stuttering detection (SD) is a difficult task. If detected at an early stage, it could facilitate speech therapists to observe and rectify the speech patterns of persons who stutter (PWS). The stuttered speech of PWS is usually available in limited amounts and is highly imbalanced. To this end, we address the class imbalance problem in the SD domain via a multibranching (MB) scheme and by weighting the contribution of classes in the overall loss function, resulting in a huge improvement in stuttering classes on the SEP-28k dataset over the baseline (StutterNet). To tackle data scarcity, we investigate the effectiveness of data augmentation on top of a multi-branched training scheme. The augmented training outperforms the MB StutterNet (clean) by a relative margin of 4.18% in macro F1-score $(\mathcal{F}_1)$ . In addition, we propose a multi-contextual (MC) StutterNet, which exploits different contexts of the stuttered speech, resulting in an overall improvement of 4.48% in $(\mathcal{F}_1)$ over the single context based MB StutterNet. Finally, we have shown that applying data augmentation in the cross-corpora scenario can improve the overall SD performance by a relative margin of 13.23% in F1 over the clean training.

Stuttering is a neuro-developmental speech impairment characterized by uncontrolled utterances (interjections) and core behaviors (blocks, repetitions, and prolongations), and is caused by the failure of speech sensorimotors. Due to its complex nature, stuttering detection (SD) is a difficult task. If detected at an early stage, it could facilitate speech therapists to observe and rectify the speech patterns of persons who stutter (PWS). The stuttered speech of PWS is usually available in limited amounts and is highly imbalanced. To this end, we address the class imbalance problem in the SD domain via a multibranching (MB) scheme and by weighting the contribution of classes in the overall loss function, resulting in a huge improvement in stuttering classes on the SEP-28k dataset over the baseline (StutterNet). To tackle data scarcity, we investigate the effectiveness of data augmentation on top of a multi-branched training scheme. The augmented training outperforms the MB StutterNet (clean) by a relative margin of 4.18% in macro F1-score (F1). In addition, we propose a multi-contextual (MC) StutterNet, which exploits different contexts of the stuttered speech, resulting in an overall improvement of 4.48% in F 1 over the single context based MB StutterNet. Finally, we have shown that applying data augmentation in the cross-corpora scenario can improve the overall SD performance by a relative margin of 13.23% in F1 over the clean training.

Stuttering is a speech disorder during which the flow of speech is interrupted by involuntary pauses and repetition of sounds. Stuttering identification is an interesting interdisciplinary domain research problem which involves pathology, psychology, acoustics, and signal processing that makes it hard and complicated to detect. Recent developments in machine and deep learning have dramatically revolutionized speech domain, however minimal attention has been given to stuttering identification. This work fills the gap by trying to bring researchers together from interdisciplinary fields. In this paper, we review comprehensively acoustic features, statistical and deep learning based stuttering/disfluency classification methods. We also present several challenges and possible future directions.

This paper discusses the way in which the classical period, such as presented in a passage from Bossuet’s Sermon sur la mort (1662), may be orally expressed. To do so, it relies on the recorded testimony left to us by Ferdinand Brunot in 1912. The first part of the paper presents the problem of sequentiality and segmentation in the classical period. The second part comments on the various manuscripts and edited versions available of the basic text. The last part of the paper compares Brunot’s analysis of his own recordings in rhetorical terms with a prosodic analysis with today’s tools. Brunot’s oralization reveals a management of pauses that favors the use of frequent and relatively long pauses, which do not allow the closing of sets, a mode of segmentation quite different from today’s practice.