Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 22 appraisals
PloS one
Pronunciation assessment in foreign language learning: Reliability and scoring bias in human-generative AI evaluation
This study examines the reliability and scoring bias of generative AI (Gen-AI)-based pronunciation assessment compared with human raters, addressing whether AI-generated scores can be trusted in real educational settings. Sixty students participated in a 12-week program. A total of 180 pronunciation samples were evaluated across eight subcomponents (individual phonemes, stress, rhythm, intonation, linking, reduction, fluency, and clarity) by three standardized human raters and Gen-AI using the same 7-point rubric. Quantitative analyses (intraclass correlation coefficients, paired-samples t-tests, and Pearson correlations) assessed reliability and bias, while semi-structured interviews with raters provided explanatory qualitative insights. Gen-AI demonstrated moderate reliability with human raters across most components, showing the highest agreement in fluency and the weakest in individual phonemes. However, Gen-AI consistently assigned significantly higher scores than human raters across all subcomponents. Qualitative findings revealed that discrepancies originated from Gen-AI's limited discriminative power, systematic flaws in handling missing data, decontextualized scoring approach, lack of sensitivity to L1 interference, and inability to interpret pragmatic context. While Gen-AI cannot fully replace human expertise, it can function as a complementary tool for formative assessment and autonomous practice. A hybrid assessment model integrating Gen-AI's efficiency with human raters' contextual and interpretive insights is recommended for effective foreign language pronunciation education.
31 July 2026
Read appraisal →The Behavioral and brain sciences
Large language models are not about natural language.
Large Language Models are useless for linguistics, as they are probabilistic models that require a vast amount of data to analyze externalized strings of words. In contrast, human language is underpinned by a mind-internal computational system that recursively generates hierarchical thought structures. The language system grows with minimal external input and can readily distinguish between real language and impossible languages.
2 July 2026
Read appraisal →Biomedical physics & engineering express
Intelligent integrated system for dysphagia rehabilitation and assessment
Current dysphagia assessment suffers from subjectivity and reliance on complex clinical tools, while rehabilitation lacks objective feedback. The critical mechanism of respiration-swallowing coordination (RSC) for preventing aspiration is often overlooked in existing monitoring systems. This study aimed to develop and validate an intelligent integrated system for dysphagia rehabilitation and that provides a closed-loop solution for quantitative assessment and personalized rehabilitation. The system's core innovation is the tight integration of a wireless wearable hardware subsystem for synchronous swallowing sound and respiratory waveform acquisition with an intelligent software subsystem. The software incorporates novel algorithms for automated swallowing identification, respiratory cycle analysis, and real-time classification of RSC patterns into aspiration risk levels. This enables an interactive platform that closes the loop between assessment, visual feedback, and personalized audiovisual guidance for rehabilitation training. Validation included hardware performance tests and a clinical trial with 16 participants (8 healthy participants and 8 participants with dysphagia). The hardware demonstrated reliable performance, achieving a robust swallowing sound SNR (14.07 dB in a noisy environment) and showing acceptable temporal consistency in the respiratory waveform. On a 0-100 scale, the system significantly differentiated swallowing function between healthy (82.72) and dysphagic (57.25) subjects (p< 0.01). We also conducted an exploratory experiment in two dysphagic participants using audiovisual guided training. A preliminary upward trend in swallowing-coordination scores was observed after training, suggesting a potential short-term training effect of the audiovisual guidance module. The integrated system represents an innovation by seamlessly merging wearable sensing with intelligent, coordination-focused algorithms into a unified closed-loop platform. It offers a practical and effective solution for transforming dysphagia management towards objective, personalized, and intelligent rehabilitation.
27 June 2026
Read appraisal →Journal of medical Internet research
Explainable and Interpretable AI for Voice and Speech Analysis in Clinical Care: Systematic Review
BACKGROUND: Driven by recent advances in artificial intelligence (AI), particularly in medicine, audio-based voice and speech biomarkers are increasingly investigated for various medical applications as a complementary or even alternative modality to traditional medical devices. The adoption of deep learning techniques in recent literature is motivated by their superior performance compared to classical machine learning methods. However, ethical and regulatory concerns regarding the black-box nature of these models have limited their integration into clinical workflows. Consequently, explainable artificial intelligence (XAI) has recently been used to address this issue by generating explanations for opaque model outputs. Ideally, medical XAI systems aim to provide human-understandable, clinically grounded explanations essential for enhanced AI trustworthiness and, thereby, facilitate adoption into real-world clinical settings. OBJECTIVE: We conduct a systematic literature review of XAI methods applied for explaining deep learning techniques in audio-based voice and speech clinical applications. We aim to identify what XAI methods have been used to explain the decisions of deep learning voice and speech AI systems in health care, as well as XAI-informed insights. Additionally, we aim to contextualize these findings with respect to clinical applicability and stakeholder relevance. Lastly, we identify opportunities and recommendations for future clinical audio XAI design. METHODS: We used PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses). Six electronic databases (IEEE Xplore, ACM Digital Library, Scopus, PubMed, Web of Science, and Nature) were searched for papers published between January 2015 and February 2025. Eligible studies applied explainability or interpretability methods to deep learning models for voice or speech audio in health care contexts. Risk of bias was assessed using PROBAST+AI (Prediction Model Risk of Bias Assessment Tool). The results were thematically synthesized across explainability categories, input representations, clinical domains, validation strategies, and stakeholder considerations. RESULTS: A total of 30 studies met the inclusion criteria. These studies used a range of explainability approaches, including gradient-based methods, perturbation-based techniques, surrogate model-based methods, model-internal representation analyses, concept-based detectors, and attention-based explanations. Applications spanned diverse clinical domains, including voice disorders, neurodegenerative diseases, psychiatric conditions, and traumatic brain injury. Overall, results indicate that most studies relied primarily on qualitative interpretation of explainability outputs, with limited quantitative validation of explanation consistency across external datasets. Furthermore, none of the included studies explicitly conducted human-in-the-loop evaluations with relevant stakeholders, highlighting a substantial gap in stakeholder alignment. CONCLUSIONS: Current XAI practices in clinical voice and speech analysis are limited by insufficient validation, lack of domain-specific design, and misalignment with clinical stakeholder needs. This review highlights opportunities for developing validated, audio-aware, and stakeholder-centered XAI approaches to support trustworthy clinical deployment. Interpretation of these findings should consider limitations related to single-reviewer study selection, potential high-risk of bias, and the repeated use of benchmark datasets.
25 June 2026
Read appraisal →Journal of speech, language, and hearing research : JSLHR
Augmentative and Alternative Communication Generative Language Intervention: A Randomized Controlled Trial
PURPOSE: Children with a wide range of impairments such as cerebral palsy and childhood apraxia of speech can benefit from augmentative and alternative communication (AAC). When AAC is provided, it frequently focuses on improving basic pragmatic skills such as requesting or on early semantic skills such as vocabulary acquisition. To address this issue, the primary goal of this study was to evaluate the effects of implementing the AAC Generative Language Intervention (AAC-GLI) program on the aided expressive grammar of preschoolers with relatively intact receptive language skills who could benefit from AAC. METHOD: A randomized controlled trial was used to evaluate the results. A total of 40 children who required AAC participated in the control group, intervention group, or both groups. The families of both the control and intervention group participants received a half-day AAC workshop. The intervention group also received 28 play-based AAC-GLI sessions. Weighted mean length of utterance in symbols (W-MLUSym), which was developed specifically for aided communicators, was used to measure progress. RESULTS: The intervention group demonstrated superior effects on W-MLUSym compared with the control group. Additionally, growth modeling demonstrated that the intervention condition was a significant moderator of change over time, with children in the intervention condition having statistically significantly greater increases in W-MLUSym, while W-MLUSym for children in the control condition remained similar to baseline. DISCUSSION: AAC-GLI can be used to teach preschoolers with a range of impairments to improve their expressive grammar skills. Providing AAC intervention that focuses on grammatical growth is an important part of expressive language development for these young children.
13 June 2026
Read appraisal →Clinical oral investigations
Exploring orofacial myofunctional therapy and its core components in mixed-dentition anterior open bite: the ombit study
OBJECTIVES: To explore the therapeutic effects of OMT in children with AOB in the mixed dentition phase, including a reduced form of OMT (focusing exclusively on breathing pattern and resting postures; OMTBPO) and traditional OMT (also including strength training and correction of swallowing; OMTBPSS). MATERIALS AND METHODS: In this double-baseline, longitudinal study, eight children with AOB (mean age 8.1 years old) were randomly allocated to the OMTBPO (n = 4) or OMTBPSS group (n = 4). Each participant received 7.5 h of OMT across 10 sessions. Outcomes included occlusal parameters, Photographic Open bite Severity Index (POSI) and Peer Assessment Rating (PAR) using 3D intraoral scans, orofacial function based on clinical and standardized assessment (Orofacial Myofunctional Evaluation with Scores; OMES), strength and endurance measurements, and oral health-related quality of life (child oral health impact profile - ortho; COHIP-ortho). RESULTS: Immediately post-therapy, anterior tongue strength had significantly increased in the full sample. At three-month follow-up, palatal volume, lateral lip endurance, and OMES function and deglutition scores had significantly increased, and vertical overbite had decreased. The PAR index showed a significant proportional reduction of 35% from baseline to the last follow-up moment (on average 6 months after therapy). The OMTBPO group showed significantly lower rates of atypical tongue posture and interdental articulation at six-month follow-up. CONCLUSION AND CLINICAL RELEVANCE: This study took a first step in exploring the effectiveness of OMT components in mixed-dentition AOB. Measurable short- to mid-term functional and occlusal improvements were found, and comparable outcomes between protocols suggest that posture-focused OMT may be an efficient interceptive approach. Further research including randomized controlled trials with larger sample sizes is required to provide evidence on efficacy and long-term stability.
1 June 2026
Read appraisal →Progress in biomedical engineering (Bristol, England)
Wearable sensors for monitoring neurogenic dysphagia: a scoping review
Neurological disorders, such as Parkinson's disease and stroke, often lead to neurogenic dysphagia, a swallowing disorder that compromises nutrition and increases the risk of malnutrition, aspiration of food, and even death. Although instrumental assessments remain the gold standard for evaluating swallowing function, their invasiveness, cost, and reliance on specialized clinical facilities limit their routine use. A systematic search was conducted in PubMed, Scopus, IEEE Xplore, and Web of Science, including only peer-reviewed studies using non-invasive wearable devices for assessing and monitoring dysphagia in neurological disorders. The initial search identified 1669 records, and an updated search retrieved 128 additional records. After screening, 24 studies met the inclusion criteria. Accelerometers and surface electromyography sensors were the most frequently used, either alone or in multimodal configurations. Multimodal approaches generally provided more accurate and comprehensive assessments than single-sensor systems. However, validation with gold-standard methods was performed in only half of the studies, and considerable heterogeneity was observed in sensor placement, study protocols, and signal-processing strategies. Wearable sensing technologies represent a non-invasive and scalable solution for assessing neurogenic dysphagia. To enable clinical translation, future research should focus on standardized methodologies, robust validation, and the integration of artificial intelligence for more precise swallowing assessment in individuals with neurological disorders.
29 May 2026
Read appraisal →Journal of neurology
A machine learning-derived speech index as a biomarker for Huntington's disease severity
BACKGROUND: The development of disease-modifying therapies for Huntington's disease (HD) necessitates sensitive, scalable, and objective biomarkers for patient stratification and tracking. Current clinical scales are rater-dependent and time-consuming, while neuroimaging is costly and inaccessible. We aim to develop and validate a novel Speech Index, derived from automated acoustic analysis, to stratify HD stages and predict the severity of disease. METHODS: We recruited 141 HTT gene carriers (37 premanifest, 104 manifest HD) and 69 healthy controls. Participants read a standardized passage, and the recordings were processed to extract 28 speech features. A composite Speech Index was constructed using Bootstrap LASSO Regression for feature selection. Its performance was validated against the HD stages, clinical scale scores and the volume of the caudate and putamen from MRI. RESULTS: The Speech Index significantly increased across HD stages (p < 0.001). The Index showed strong correlations with clinical measures (cUHDRS: ρ = -0.67; TMS: ρ = 0.57; SDMT: ρ = -0.63; all p < 0.001) and neuroimaging biomarkers (caudate: ρ = -0.55; putamen: ρ = -0.62; all p < 0.001). Generalized additive models confirmed the Index's high predictive value for these outcomes (pseudo-R2 from 0.430 to 0.596). CONCLUSION: We developed a fully automated, interpretable Speech Index that serves as a valid digital biomarker for HD severity. It holds promise for remote monitoring, clinical trial enrichment, and objective assessment of therapeutic efficacy.
23 May 2026
Read appraisal →Journal of medical Internet research
Efficacy of Digital Speech Therapy for Poststroke Dysarthria: Randomized Noninferiority Trial
BACKGROUND: Poststroke dysarthria, a common speech impairment, affects up to half of all stroke survivors, often reducing their ability to communicate, and adversely affecting their quality of life. Although conventional speech therapy for poststroke dysarthria is effective, access is often limited by time and geographical constraints. Here, digital speech therapy may serve as a remotely deliverable alternative for selected patients. However, few trials have assessed its efficacy, safety, and usability. OBJECTIVE: This study aimed to evaluate whether a smartphone-based speech therapy app is noninferior to conventional workbook-based therapy in improving speech intelligibility among individuals with poststroke dysarthria. METHODS: This single-blind, randomized controlled, noninferiority trial was performed at 3 hospitals in South Korea. Adults (≥19 y) with poststroke dysarthria who were cognitively intact, without aphasia, and able to use a smartphone were eligible. Participants were enrolled between July 20, 2023, and April 15, 2024. Participants were randomly assigned (1:1), stratified by stroke phase, using a block randomization method, to receive either a smartphone-based digital therapy app or a conventional workbook-based therapy for 4 weeks. The primary outcome was speech intelligibility (0-100 perceptual rating) after the intervention. Primary analysis was intention-to-treat using analysis of covariance. A noninferiority margin of 19 points was pre-defined. RESULTS: A total of 73 participants were enrolled (median age 62.00 years). Among them, 38 were assigned to the digital speech therapy group and 35 to the control group. Intelligibility scores improved from 80.48 (SD 18.92) to 92.08 (SD 12.38) in the intervention group, and from 80.94 (SD 16.74) to 88.11 (SD 18.06) in the control group. The adjusted between-group difference was 4.49 (95% CI 0.61-8.37), and the lower bound of the 95% CI was above the prespecified noninferiority margin (-19), which supported noninferiority. No significant between-group differences were observed in the secondary outcomes related to speech function or psychological status. The system usability score was 89.6, and adherence in the digital speech therapy group was 64.6% based on app logs, with no treatment-related adverse events. CONCLUSIONS: Digital speech therapy was noninferior to conventional workbook-based therapy in improving speech intelligibility and was feasible across acute to early subacute and chronic stroke phases in cognitively intact stroke survivors with predominantly mild-to-moderate dysarthria. However, feasibility and efficacy in older stroke survivors with cognitive deficits or co-occurring aphasia, or in those unable to use smartphones, remain to be established.
19 May 2026
Read appraisal →PloS one
Evaluation of the diagnostic value of the Modified Evan's Blue Dye Test for assessing aspiration in tracheostomized critically ill patients: A systematic review and meta-analysis
Dysphagia and aspiration are common complications in tracheostomized critically ill patients. The Modified Evan's Blue Dye Test (MEBDT) is a non-invasive, bedside, adjunctive rule-in signal for aspiration when positive. However, its diagnostic accuracy compared to the gold standard Fiberoptic Endoscopic Evaluation of Swallowing (FEES) in these patients remains unclear. We conducted a systematic review and meta-analysis to evaluate the diagnostic accuracy of the MEBDT for the detection of aspiration in tracheostomized critically ill patients. Study protocol was prospectively registered on PROSPERO (CRD:42023479920). A systematic search was performed across three databases (PubMed, Cochrane Central, and Embase), followed by a systematic screening process against predetermined selection criteria and included studies that provided data on the sensitivity and specificity of MEBDT versus FEES. The risk of bias and the level of evidence certainty in the included papers were assessed by the QUADAS-2 tool and GRADE approach respectively. Six out of 2227 screened studies were included. We found that MEBDT had a high specificity 95.42% (95% CI [67.38%, 99.53%]) and positive predictive value 95% (95% CI [81, 100]). In conclusion, the MEBDT can serve as a bedside adjunctive tool that indicates a potential aspiration risk when the result is positive. However, if the result is negative, further diagnostic assessments, like FEES, are recommended for high-risk patients.
15 May 2026
Read appraisal →Proceedings of the National Academy of Sciences of the United States of America
Discovering regularity and mechanisms of word sense acquisition in childhood.
How does language use inform the emergence of word meanings in early life? Prior work in developmental psychology and the cognitive sciences typically focuses on studying word acquisition in children without specifying how different senses within a word emerge through time. To shed light on word sense acquisition, we propose a framework grounded in state-of-the-art computational methodologies of contextual word embedding to characterize how different senses of a word unfold as children acquire their lexicon. Our framework identifies word senses automatically by forming semantic clusters through natural language use, and analyzes 1,270 words from approximately 4 million utterances produced by children (19 to 144 mo) and their caretakers. The psychological validity of these senses was assessed based on a combination of dictionaries, human judgment of semantic similarity, and evaluation from a large language model. We then tested three hypotheses motivated by existing work on word sense emergence in language evolution. First, concrete senses of a word tend to emerge earlier than more abstract senses in child language. Second, word senses grow incrementally in semantic space across development. Third, algorithms of semantic chaining-how words spawn new senses by extending from existing senses-recapitulate the order of word senses in development. We find support for all three hypotheses. Our work suggests that the development of word meaning in language acquisition resembles that in language evolution and offers a converging view on the cognitive principles and mechanisms across timescales in the ontogeny and phylogeny of word sense emergence.
13 May 2026
Read appraisal →Journal of speech, language, and hearing research : JSLHR
Vocal Biomarkers of Childhood Trauma: A Machine-Learning Approach to Speech Analysis
PURPOSE: Childhood trauma can disrupt communication, yet early signs often go unrecognized in regions affected by ongoing war, where immediate physical needs take precedence. Vocal biomarkers-acoustic features linked to emotional and motor regulation-offer a promising, noninvasive means of detecting trauma-linked speech disruptions. This study applied a hybrid framework to distinguish trauma exposure in Arabic-speaking children living amid active conflict. The aim was to support scalable, speech-based tools for early trauma identification in low-resource, humanitarian settings. METHOD: We analyzed 200 publicly available recordings of spontaneous speech from Arabic-speaking girls (ages 8-12 years): 100 trauma-exposed participants from Gaza (Palestinian) and 100 non-exposed controls from Jordan. Core acoustic features (fundamental frequency [F0], jitter, shimmer, harmonics-to-noise ratio [HNR], voice onset time [VOT], first formant, second formant) informed statistical testing and theory-driven composite indices. Exploratory features-including Mel-frequency cepstral coefficients and eGeMAPSv02 descriptors-were used to train binary classification models. Three classifiers (random forest, ridge regression, and logistic regression) were evaluated using nested cross-validation and bootstrap resampling. Composite indices were combined into a 0-10 Trauma Risk Score. Generalizability was assessed using an independent Lebanese cohort (n = 80), with the trained classifier applied using fixed exploratory features and preprocessing parameters. Core features were tested post hoc for cross-cohort stability. RESULTS: Trauma-exposed children showed reduced F0 and HNR, elevated shimmer and jitter, and prolonged VOT (Cohen's d > 1.2). Binary classification models achieved strong performance (area under the curve [AUC] = .89-.92); logistic regression reached AUC = .996 under cross-validation. Composite indices (AUCs > .90) stratified 68% into Moderate/High Trauma Risk. Lebanese validation confirmed generalizability, with theory-driven features showing stable predictive patterns. CONCLUSIONS: Vocal biomarkers reliably distinguished trauma exposure in Arabic-speaking children using a simple logistic regression model. This strong performance highlights the potential of speech-based tools as scalable, noninvasive methods for early trauma detection. Further validation is needed to support their use in diverse humanitarian and conflict-affected settings.
12 May 2026
Read appraisal →American journal of speech-language pathology
Parental Roles in Language Intervention: A Systematic Review
BACKGROUND: Children with language delays (LDs) or language disorders face persistent challenges in their development. Speech and language therapy is vital, and within it, parental involvement is recognized as a key factor in therapeutic success. Although many parent-implemented interventions exist, little is known about the specific roles parents assume, particularly across different diagnostic groups: LD, developmental language disorder (DLD), and language disorder associated with other conditions (LD + X). OBJECTIVE: The aim of this study was to review and synthesize evidence on parental roles in language interventions for children with LD, DLD, and LD + X. METHOD: Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and registered in PROSPERO (CRD42023452616), this systematic review analyzed studies from four databases. Eligible studies involved parent participatory language interventions reporting child language outcomes. Two independent reviewers conducted screening, quality appraisal, and data extraction. RESULTS: Across 37 studies, parents most frequently assumed the roles of Learner, Implementer, and Informer, highlighting their functions in acquiring knowledge, applying strategies, and providing feedback on children's progress. Less common roles included Adaptor and Collaborative Decision-Maker, suggesting that although caregivers are increasingly active, their engagement rarely extends to shared decision making with professionals. The studies reviewed here demonstrated considerable variability in intervention dosage, total duration, delivery format, and strategies taught. In contrast, less variability was observed in the identity of the parents/caregivers (most often the mother) and the trainer (typically a speech-language therapist) as well as in the type and mode of delivery. CONCLUSIONS: Parental engagement in language interventions continues to be largely determined by professional guidance, positioning parents as learners and implementers rather than full collaborators. Strengthening adaptive roles and promoting shared decision making can enhance family-centered practice. Future programs could consider the specific needs of each context, encourage parental involvement, and explore telepractice to expand access. SUPPLEMENTAL MATERIAL: https://doi.org/10.23641/asha.32060328.
11 May 2026
Read appraisal →American journal of speech-language pathology
Effect of the Fake Phone Call App on Patient Adherence Behavior and Motivation in Conversation Training Therapy: A Randomized Crossover Trial
PURPOSE: The Fake Phone Call app was previously developed and feasibility-tested for conversational target voice practice in resonant voice therapy. The purpose of the present study was to examine its effect on adherence and motivation for conversational practice in conversation training therapy (CTT) as well as its usability and utility in CTT. METHOD: Eleven patients referred for voice therapy served as their own control in a randomized crossover trial comparing 1 week of app-supported homework practice with 1 week of standard-of-care homework completion in CTT. Dependent variables included adherence and motivation variables. In addition, usability and utility as well as outcomes were examined at study completion. RESULTS: App support significantly increased practice frequency, practice fidelity, and goal commitment for practice. Usability and utility were good. Exploratory investigation revealed that patients randomized to receive the app condition first and the standard-of-care condition second had greater gains on all study variables and significantly greater self-reported generalization at treatment completion. CONCLUSION: Fake Phone Call support significantly improves adherence and motivation for CTT, with additional benefits when provided earlier in therapy.
11 May 2026
Read appraisal →BMJ health & care informatics
AI-generated clinical summaries: errors and susceptibility to speech and speaker variability
OBJECTIVES: To evaluate whether variability in patients' communication style (personality), international English-accents (human and synthetic) and speech impairments affects the accuracy of a Clinical AI Scribe (CAIS) and identify where performance degrades to inform pre-deployment validation and monitoring. METHODS: We conducted simulated primary-care consultations using trained actors. For personality types, four scenarios were enacted, each with five patient-personality types. For accents, transcripts of consultations were used to generate combinations of seven accents across five scenarios. The CAIS produced summaries that were compared with transcripts, and errors classified as omissions, factual inaccuracies or hallucinations. For speech impairments, public recordings representing five profiles were transcribed and word-recognition accuracy calculated. RESULTS: Personality types showed no statistically significant differences in errors (all p>0.05). Extraversion had the highest total errors (median 3.5). Across accents, comparisons were non-significant for both patient and doctor voices (patients: p=0.851; doctors: p=0.980). Omissions predominated, with low rates of hallucinations and factual inaccuracies. Omissions were slightly higher for Chinese-accented and Indian-accented doctors (both medians 3.0). Conversely, speech impairments differed: cleft palate and vowel disorders were near-perfect, whereas phonological impairment markedly reduced recognition (p<0.001). DISCUSSION: Operationally, CAIS deployment should include clinician-in-the-loop verification, subgroup performance monitoring (accents, impairments) and predefined 'switch-off' criteria for severe phonological patterns. High-quality synthetic voices are a pragmatic proxy for accent testing when balanced corpora are unavailable. CONCLUSIONS: Under controlled conditions CAIS performance was broadly stable across communication styles and most accents, but vulnerable to specific speech characteristics, particularly phonological impairment, in this single-system simulation study.
27 Apr 2026
Read appraisal →PloS one
Comparison of deep learning-based three-dimensional human pose estimation methods with motion capture for gesture research
Humans often make gestures while speaking. These gestures have been extensively researched in psychology and cognitive science, revealing their functions and roles in communication. Despite being produced in three-dimensional (3D) space, gestures and gesture space have primarily been measured and analyzed using two-dimensional planes. Optical motion capture (MoCap) is commonly used for 3D gesture measurement. However, MoCap is expensive, requires a large space, and the attachment of retroreflective markers can interfere with the natural generation of gestures by the speaker. Recent advances in deep learning-based human pose estimation (HPE) offer promising alternatives by enabling 3D keypoint estimation from standard video cameras. This study investigates the accuracy of four HPEs-two monocular and two stereo with triangulation methods-in estimating upper-body keypoints commonly used in natural scenarios for gesture research. Ten participants were recorded performing gesture-rich speech by both a MoCap system and three video cameras. We compared 13 keypoints, including the wrists, elbows, shoulders, fingers, and face, with MoCap data, using Euclidean distance as an error metric. Statistical analyses revealed that stereo methods significantly outperformed monocular methods across all keypoints. The most accurate method achieved an average error of 49.4 mm with respect to MoCap, suggesting sufficient accuracy for practical gesture analysis. Additionally, we visualized 3D gesture spaces and found a 75.4% overlap between HPE and MoCap at a voxel size of 50 mm, indicating high spatial agreement. Our findings demonstrate that stereo HPE methods are viable, cost-effective alternatives to MoCap in gesture-related applications. Our ultimate objective is to create a toolbox that is accurate and easy to use for measuring human gestures in 3D space by exploiting the recent advances in HPE as an alternative to MoCap. Such a toolbox will be suitable for non-experts in machine learning, and these results lay the foundation for it.
27 Apr 2026
Read appraisal →Biomedical physics & engineering express
A multimodal attention fusion-based model for pathological voice detection
Pathological voice detection provides a non-invasive and cost-effective approach for the early screening of laryngeal disorders. However, most existing methods rely on single-modal acoustic features and mainly focus on binary classification, which limits their ability to achieve fine-grained discrimination among multiple pathological voice categories. To address these limitations, this paper proposes a Multimodal Fusion Network (MFNet) with attention-based fusion for pathological voice classification. The proposed framework jointly models time-domain and frequency-domain information within a unified architecture. Specifically, SincNet is employed to extract task-relevant narrowband acoustic features directly from raw speech waveforms, while the Tunable Q-factor Wavelet Transform (TQWT) and SKNet are combined to learn multi-scale spectral representations from Mel-spectrograms. In addition, an Attentional Feature Fusion (AFF) module is introduced to adaptively integrate features from the two branches, thereby enhancing the discriminative capability of pathological voice representations. Comparative experiments on three public pathological voice datasets show that the proposed method outperforms representative baseline models and demonstrates strong robustness and generalization across different datasets. Furthermore, ablation studies further validate the effectiveness of the proposed architectural design. These results indicate that MFNet provides an effective solution for pathological voice classification and has promising potential for computer-aided voice disorder screening.
18 Apr 2026
Read appraisal →Journal of speech, language, and hearing research : JSLHR
Information Load Predicts Stuttering Events
PURPOSE: One potential contributor to the variability of stuttering events is information load, quantified by assessing word predictability based on preceding context. This study investigated whether words with greater information load are more likely to be stuttered using word surprisal as an index while controlling for relevant linguistic variables. We also examined whether this relationship varied with stuttering severity and impact. METHOD: We analyzed spontaneous speech samples from 35 adults who stutter, drawn from the FluencyBank corpus. Word surprisal values were computed using the Generative Pretrained Transformer 2 large language model and analyses controlled for six linguistic predictors of stuttering. We also tested whether clinical measures of stuttering severity and overall impact predicted individuals' sensitivity to surprisal. RESULTS: Higher word surprisal significantly increased the likelihood of stuttering, even after controlling for linguistic covariates. However, we found no evidence that stuttering severity or overall impact moderated this relationship. CONCLUSIONS: These findings suggest that words with greater information load are more likely to elicit stuttering during spontaneous speech. This pattern may reflect increased communicative demands associated with high-surprisal words. The absence of a moderating effect of stuttering severity or impact may indicate that individual differences in coping with communicative pressure, rather than clinical severity alone, shape stuttering variability.
11 Apr 2026
Read appraisal →Journal of speech, language, and hearing research : JSLHR
Neuroimaging Advances in Brain Function of Laryngeal Dystonia: A Systematic Review
PURPOSE: This study aims to review neuroimaging research on brain function in laryngeal dystonia (LD), summarize current evidence on its pathophysiological mechanisms, and highlight directions for future investigation. METHOD: PubMed, Scopus, and Web of Science Core Collection were searched from database inception to November 2025. Two reviewers independently screened studies, extracted data, and assessed study quality using the Newcastle-Ottawa Scale. Eligible studies were synthesized through systematic review and visual analysis. RESULTS: A total of 33 studies were included. These studies employed imaging techniques such as functional magnetic resonance imaging and voxel-based morphometry to identify abnormalities in patients with LD. Structurally, key findings include gray matter alterations in sensorimotor areas, reduced thalamic volume, and impaired white matter integrity. Functionally, this manifests as sensorimotor integration dysfunction (characterized by M1/S1 and cerebellar hypoactivation alongside pallidal and thalamic hyperactivation) and network decoupling (e.g., weakened M1-S1 connectivity). CONCLUSIONS: The pathological mechanism of LD is a complex network imbalance involving the core "cortico-basal ganglia-thalamo-cerebellar" loop, driven primarily by sensorimotor integration dysfunction. Current research is limited in quantity and suffers from significant methodological heterogeneity. Future studies should further clarify the functional interactions among brain regions and their relationship with clinical symptoms, thereby providing a theoretical basis for the development of targeted therapies. SUPPLEMENTAL MATERIAL: https://doi.org/10.23641/asha.31329103.
11 Apr 2026
Read appraisal →Journal of speech, language, and hearing research : JSLHR
Rhotic Acquisition Is More Rapid in Biofeedback Than Motor-Based Treatment for Residual Speech Sound Disorder: Primary Outcome of a Randomized Controlled Trial
PURPOSE: Residual speech sound disorder (RSSD) is a high-prevalence condition that can limit children's academic and social participation, with negative consequences for overall well-being. Previous studies have described visual biofeedback as a promising option for RSSD, but results have been inconclusive due to study design limitations and small sample sizes. METHOD: In a preregistered randomized controlled trial, 108 children aged 9-15 years with RSSD affecting American English /ɹ/ were randomly assigned to receive treatment incorporating visual biofeedback (subdivided into ultrasound and visual-acoustic types) or a comparison condition of motor-based treatment consistent with current best practices in speech therapy. An acoustic measure (the distance between the second and third formants, which is smaller in perceptually accurate /ɹ/) was used to quantify progress in the first three sessions of treatment in participants' assigned condition. The focus on the early stages of treatment was grounded in principles of motor learning: As a type of knowledge of performance feedback, biofeedback is expected to have its greatest impact during initial acquisition of a new speech-motor plan. RESULTS: A linear mixed-effects regression model revealed a statistically significant interaction between treatment type and session: While both groups made progress over time, the rate of progress was significantly faster in the biofeedback condition compared to motor-based treatment. No statistically significant difference was observed between the ultrasound and visual-acoustic types of biofeedback. DISCUSSION: This large-scale randomized controlled trial supports previous small-scale research in finding that biofeedback can help children with RSSD acquire American English /ɹ/. A companion study will compare participants' generalization learning after the end of all treatment. SUPPLEMENTAL MATERIAL: https://doi.org/10.23641/asha.29971501.
11 Apr 2026
Read appraisal →Journal of neural engineering
Large-scale training data enhances silent speech decoding with around-ear EEG.
Objective. Silent speech decoding (SSD) offers a potential communication alternative for individuals with impaired vocalization. However, conventional multi-electrode electroencephalography (EEG) or facial electromyography (EMG) systems require cumbersome preparation and are unsuitable for daily use. This study evaluates the practicality of SSD using a wearable around-ear EEG device, focusing on data scaling, cross-subject transfer, vocabulary extensibility, and online decoding performance.Approach. We collected 72 h of around-ear EEG from 24 healthy participants and one individual with incomplete locked-in syndrome (LIS) during silent, vocalized, and attempted speech, and integrated these around-ear EEG recordings with prior EMG + high-density EEG datasets, yielding 282.4 total h of training data. Using a 64-word classification task as the evaluation metric, we assessed: (1) whether larger datasets improve around-ear EEG-based SSD, (2) whether healthy-participant data supplement limited LIS-participant data despite articulatory differences, (3) transferability to unseen vocabulary, and (4) online user-interface performance.Main results. Large-scale EEG/EMG data improved SSD accuracy in both healthy participants and the LIS participant. Training on the heterogeneous dataset achieved 56.6% accuracy for healthy users and 47.3% for the LIS participant. Fine-tuning this decoder for new vocabulary increased the accuracy by 22 percentage points relative to training from scratch. Regression analysis showed that, for decoding in the LIS participant, data from the LIS participant contributed approximately four times the weight of healthy-participant data, quantifying data strategies for SSD. Online experiments achieved top-1/top-5 accuracies of 47.2%/76.0% for healthy users and 26.5%/49.1% for the LIS participant.Significance. The results indicate that lightweight, commercially feasible around-ear EEG can enable practical SSD when combined with large-scale healthy-participant data, supporting online operation. Moreover, models trained on a 64-word vocabulary facilitate decoding of a new vocabulary, providing a path toward SSD systems requiring minimal LIS-participant data. This study advances non-invasive SSD systems suitable for everyday communication.
8 Apr 2026
Read appraisal →Advanced science (Weinheim, Baden-Wurttemberg, Germany)
A Wireless, Battery-Free Artificial Throat Patch with Deep Learning for Emotional Speech Recognition
Technologies for sensing and translating throat signals into speech have advanced the communication for people with vocal impairments. However, the practical deployment of these technologies is currently hindered by tethered system, bulky components and the lack of emotional recognition capability. Here, we demonstrate a wireless, battery-free artificial throat patch system (ATPS) coupling with deep learning method to enable simultaneous recognition of speech and emotion. The sensing module of the ATPS integrates a carbon nanotube-based thin-film strain sensor and a miniaturized flexible printed circuit board (FPCB). Real-time sensing and transmission of throat signals is facilitated by a smartphone-linked system incorporating an FPCB with an embedded near-field communication antenna and low-power electronic components. Furthermore, the information-rich sensor readouts are further extracted and successfully demonstrated to be sufficient to emotional recognition of speech by implementing a hybrid deep learning architecture. This proposed system expands the arsenal of tools available for mute patients in speech recognition and communication and further realizes personalized clinical implementations.
4 Apr 2026
Read appraisal →