Search Results (237)

This paper presents the first prototype of a speech-to-sign language translation system for Kazakh Sign Language (KRSL). The proposed pipeline integrates the NVIDIA FastConformer model for automatic speech recognition (ASR) in the Kazakh language and addresses the challenges of sign language translation in a low-resource setting. Unlike American or British Sign Languages, KRSL lacks publicly available datasets and established translation systems. The pipeline follows a multi-stage process: speech input is converted into text via ASR, segmented into phrases, matched with corresponding gestures, and visualized as sign language. System performance is evaluated using word error rate (WER) for ASR and accuracy metrics for speech-to-sign translation. This study also introduces the first KRSL dataset, consisting of 1200 manually recreated signs, including 95% static images and 5% dynamic gesture videos. To improve robustness under resource-constrained conditions, a Weighted Hybrid Similarity Score (WHSS)-based gesture matching method is proposed. Experimental results show that the FastConformer model achieves an average WER of 10.55%, with 7.8% for isolated words and 13.3% for full sentences. At the phrase level, the system achieves 92.1% accuracy for unigrams, 84.6% for bigrams, and 78.3% for trigrams. The complete pipeline reaches 85% accuracy for individual words and 70% for sentences, with an average latency of 310 ms. These results demonstrate the feasibility and effectiveness of the proposed system for supporting people with hearing and speech impairments in Kazakhstan. Full article

STT3A encodes the catalytic subunit of the oligosaccharyltransferase A (OST-A) complex and is classically linked to severe autosomal-recessive congenital disorder of glycosylation (CDG). To define the distinct autosomal-dominant disorder, we reviewed all published cases and integrated three previously unpublished individuals from the CDG natural history study. Across 21 individuals, abnormal transferrin glycosylation was present in nearly all individuals (20/21), and subtle facial dysmorphism was common (18/21). Neurodevelopmental involvement was frequent, including motor delay (13/21), learning difficulties (13/21), speech delay (12/21), and intellectual disability (10/21). Musculoskeletal manifestations were also common, including skeletal abnormalities (12/21), short stature (11/21), muscle cramps (8/21), and early-onset osteoarthritis in adults (6/21). Less frequent features included congenital heart defects (5/21) and coagulation factor deficiency (5/21). Importantly, the newly reported individuals expand dominant STT3A-CDG with previously unreported features, including anorectal malformation, morbid obesity, and clinically significant bleeding diathesis with von Willebrand factor and factor VIII deficiency. Biochemical signatures ranged from classic type I transferrin patterns to subtle or atypical abnormalities, emphasizing that near-normal transferrin testing does not exclude the diagnosis. Variants clustered in conserved catalytic regions, with recurrent p.Arg405 across de novo, inherited, and mosaic cases supporting a mutational hotspot and likely dominant-negative mechanism. Full article

Developmental Language Disorder (DLD) is a common neurodevelopmental condition that affects language acquisition in children. However, standardized diagnostic tools for Tunisian Arabic, a widely spoken yet underrepresented dialect, is still lacking. This study presents a multimodal biomedical informatics framework that integrates clinical assessments, speech recordings, and artificial intelligence (AI) for early DLD detection. Three linguistic tasks (the CLT Task, the Arabic Verb Evaluation Task, and the Nonword Repetition Task) were adapted for Tunisian Arabic, and spontaneous speech samples were collected from children with typical development and those with DLD. Statistical analyses revealed significant deficits in verb production, past-tense morphology, and phonological memory in the DLD group. For automated screening, we developed two systems: a Random Forest classifier based on structured clinical and linguistic features and a multimodal deep learning model using Wav2Vec2 acoustic embeddings. The best model achieved an F1 score of 0.85, demonstrating the feasibility of AI-assisted DLD screening. This work introduces the first standardized dataset and computational baseline for DLD in Tunisian Arabic, providing clinically relevant tools for early identification and supporting research on underrepresented Arabic dialects. This work also highlights future implications, including potential applications in early screening, the integration of acoustic markers, and the development of culturally adapted assessment tools for underrepresented languages. Full article

A persistent challenge for Deaf and Hard-of-Hearing individuals is the communication gap between sign language users and the hearing community, particularly in regions with limited automated translation resources. In Saudi Arabia, this gap is amplified by the reliance on Saudi Sign Language (SSL) and the scarcity of real-time, sentence-level translation systems. This paper presents a real-time system for sentence-level recognition of continuous SSL and direct mapping to natural spoken Arabic. The proposed system operates end-to-end on live video streams or pre-recorded content, extracting spatio-temporal landmark features using the MediaPipe Holistic framework. For classification, the input feature vector consists of 225 features derived from hand and body pose landmarks. These features are processed by a Bidirectional Long Short-Term Memory (BiLSTM) network trained on the ArabSign (ArSL) dataset to perform direct sentence-level classification over a vocabulary of 50 continuous Arabic sign language sentences, supported by an idle-based segmentation mechanism that enables natural, uninterrupted signing. Experimental evaluation demonstrates robust generalization: under a Leave-One-Signer-Out (LOSO) cross-validation protocol, the model attains a mean sentence-level accuracy of 94.2%, outperforming the fixed signer-independent split baseline of 92.07%, while maintaining real-time performance suitable for interactive use. To enhance linguistic fluency, an optional post-recognition refinement stage is incorporated using a large language model (LLM), followed by text-to-speech synthesis to produce audible Arabic output; this refinement operates strictly as post-processing and is not included in the reported recognition accuracy metrics. The results demonstrate that direct sentence-level modeling, combined with landmark-based feature extraction and real-time segmentation, provides an effective and practical solution for continuous SSL sentence recognition in real-time. Full article

Accurate Mispronunciation Detection and Diagnosis (MDD) for non-native Korean learners is critical for effective pronunciation feedback, but it is hindered by the lack of training labels that reflect learners’ actual pronunciations. This paper presents a pseudo-label generation framework that fine-tunes Whisper to output pronunciation-oriented sequences, supported by data-quality management and iterative label refinement. We convert orthographic transcripts into pronunciation targets using an existing Grapheme-to-Phoneme (G2P) tool to reduce reliance on standard written forms, and apply multi-stage refinement with cross-model agreement validation under progressively adjusted thresholds to filter unreliable pseudo-labels. We further improve robustness by incorporating larger and more diverse non-native speech corpora and by applying dataset-specific preprocessing, including noise removal, duration-based selection, and duplicate control. Evaluation on a manually annotated test set of actual learner pronunciations shows that models trained with refined pseudo-labels achieved a lower Phoneme Error Rate (PER) and performed better than the baseline model on MDD. Overall, the proposed framework enables practical MDD for non-native Korean speech without requiring large-scale manual phoneme annotation. Full article

Background/Objectives: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that affects the daily lives of older adults, impacting their cognitive abilities as well as speech and language communication. Early detection is crucial, as it enables timely intervention and helps improve the quality of life for those affected. While large language models (LLMs) have shown promise from spontaneous speech, most studies are unimodal and miss complementary signals across modalities. Methods: We present an LLM-powered multimodal cross-attention framework that integrates lexical (text), acoustic (speech), and visual (image) information for dementia detection using the ADReSSo 2021 picture-description dataset. Within this framework, text data are encoded using the ModernBERT, audio features are extracted using the wav2vec 2.0-base-960, and the Cookie Theft image is represented through the CLIP ViT-L/14. These embeddings are linearly projected to a shared space and then combined via Transformer-based cross-attention, yielding a fused vector for AD detection. Results: Our results show that the trimodal model achieved the best overall performance when paired with an SVC classifier, reaching an accuracy of 0.8732 and an F1 score of 0.8571, surpassing both the top-performing unimodal and bimodal configurations. For interpretability, a sensitivity analysis of modality contributions reveals that text plays the primary role, audio provides complementary improvements, and image offers modest yet stabilizing contextual support. Conclusions: These results highlight that the method of multimodal embedding fusion significantly influences performance: a cross-attention block achieves an effective balance between accuracy and simplicity, producing integrated representations that align well with interpretable downstream classifiers. Full article

Generative AI offers transformative potential for scalable, personalized, and dynamic language education, particularly in enhancing oral proficiency among young learners. However, effective deployment remains challenging due to limited resources for some languages, the need for age-appropriate content and tools, and the importance of respecting cultural relevance. In this paper, we introduce LEARN (Language Evaluation via question Answer generation from caRtooNs), a culturally grounded multilingual visual dialogue system designed to support oral proficiency in three of Singapore’s official languages: Mandarin, Bahasa Melayu, and Tamil. English, as the lingua franca, is excluded. LEARN integrates a teacher-facing module for curriculum-aligned visual question-answering task creation and a student-facing module for voice-driven adaptive dialogue, optimized for children’s speech. Unlike existing platforms, LEARN prioritizes cultural relevance and low-resource language support, helping address gaps in heritage language preservation. Pilot studies with students demonstrate significant improvements in engagement and vocabulary acquisition. Designed for classroom as well as home use, LEARN presents a scalable AI-driven language tutoring framework. Full article

Background/Objectives: Locally advanced midface malignant tumors require extensive resection, resulting in complex defects involving bone and multiple soft tissue structures. Reconstructing these substantial defects presents a significant challenge to restore both function and aesthetics. This study aims to compare the functional and aesthetic outcomes of chimeric free flaps versus single free flaps in midface microvascular reconstructions. Methods: This retrospective analysis included fifty consecutive patients with Type III Cordeiro defects who underwent midface reconstruction with free tissue transfer between 2020 and 2024. The cohort included fourteen patients who received prefabricated chimeric flaps and thirty-six patients who received single free flaps. Outcomes were assessed six months postoperatively using a modified University of Washington Quality of Life Questionnaire (UW-QOL), analyzing domains including speech, chewing, sensation, appearance, pain, and social activity. Statistical analysis was performed using the Mann–Whitney U test. Results: In the chimeric flap group, no major flap necrosis or complications were observed. In unadjusted comparisons, the chimeric flap group showed higher transformed UW-QOL scores in several domains. Statistically significant between-group differences were observed for opening and speech (p = 0.004), change in appearance (p = 0.022), sensation (p = 0.011), and social activity (p = 0.006). Aesthetic outcomes, assessed via patient rating of appearance, were also significantly higher in unadjusted comparisons with the chimeric flap approach. Furthermore, in Type IIIa defects, titanium mesh successfully provided reliable orbital support. Conclusions: Chimeric free flaps represent a feasible reconstructive option in selected cases of complex maxillary and midface reconstruction. Their main advantages—providing the proper amount of specific, well-vascularized tissue and offering greater mobility of components— may be associated with more favorable functional, aesthetic, and social outcomes in unadjusted comparisons compared to reconstruction using single free flaps. Full article

Early life experiences and the process of exploration play a vital role in shaping brain development and lifelong learning. In March 2020, population-wide restrictions were imposed due to the COVID-19 pandemic. It remains to be determined whether having been raised under the global stress and restrictions of COVID-19 has influenced children’s development as they enter formal schooling. The aim of this study was to examine the extent to which having more than 50% of one’s first year of life and/or prenatal period in the COVID-19 era influences the developmental trajectory in preschool. The study compared 3- to 5-year-old children born before the pandemic (n = 63) with those who were five months or younger at its onset (n = 40). Variables assessed included executive function skills, vocabulary, and common developmental domains. Using the BRIEF-P as a standardized measure of executive function, the results demonstrate that the pandemic-born cohort exhibit greater impairments than those born before the pandemic. There was also a significant increase in reports of speech and language therapy enrollment; frequent ear infections; diagnoses of hearing, speech, or language impairments; and delays in reaching developmental milestones. The pandemic-born cohort additionally reported delays in fine motor skills compared to the pre-pandemic cohort. The present study underscores the urgent need for additional resources to better support children in this cohort as they begin formal schooling. Full article

This article presents AgoraAI, an open-source framework designed to enable dynamic, multi-participant conversations by integrating Multi-Persona Orchestration within a shared conversational environment. Unlike traditional single-agent Large Language Model (LLM) interactions or passive commercial meeting assistants, AgoraAI allows users to configure distinct AI personas that engage in active facilitation and simultaneous, turn-based dialogues with human participants. The system supports diverse high-stakes use cases, including formal panel discussions and interactive educational settings. Crucially, this work addresses the engineering challenge of the “Concurrency-Coherence Paradox” in real-time voice systems. Key architectural contributions include: (1) the implementation of Asynchronous Dual-Queue Processing, a thread-safe integration strategy that synchronizes real-time Speech-to-Text streams with LLM generation to resolve race conditions; and (2) Dynamic Context-Injection pipelines that ensure persona consistency. The platform’s ecological validity is demonstrated through deployment in a human-supervised Master’s thesis seminar and a corporate coordination meeting. Results from an exploratory pilot study indicate high usability, perceived utility, and strong user acceptance. These findings suggest that AgoraAI provides a flexible, empirically evaluated architecture for democratizing multi-perspective collaboration across education, research, and professional domains. Full article

Background/Objectives: Speech understanding in noise relies on both temporal fine structure (TFS) and temporal envelope (ENV) cues. While TFS primarily conveys interaural time differences (ITDs) at low frequencies, ENV cues can also support ITD processing, especially when TFS is unavailable or degraded. Expanding the ENV by increasing modulation depth has been proposed to improve speech perception, but its effects on spatial release from masking (SRM) and binaural temporal processing in normal-hearing listeners remain unclear. The goal of this study was to evaluate the effect of ENV enhancement on SRM in young adults with normal hearing and its influence on ITD sensitivity and interaural coherence (IC). Method: Thirty normal-hearing native Kannada speakers (19–34 years) participated. Speech stimuli consisted of Kannada sentences embedded in four-talker babble at −5, 0, and +5 dB signal to noise ratio (SNR). Target and masker were spatialized using head-related transfer functions at 0°, 15°, and 37.5° azimuths. Stimuli were presented with and without ENV enhancement (compression–expansion algorithm). Speech recognition scores were analyzed using generalized linear mixed models, and SRM was calculated as performance differences between co-located and spatially separated conditions. Cross-correlation analyses were performed to estimate ITDs and IC across SNRs. Result: ENV enhancement yielded significantly higher SRM values across all SNRs and spatial separations. Benefits were greatest at lower SNRs and wider target–masker separations. Cross-correlation analysis showed enhanced IC and more reliable ITD estimates under the expanded condition, particularly at moderate SNRs. Conclusions: Temporal ENV enhancement strengthens spatial unmasking and binaural timing cues in normal-hearing adults, especially under adverse listening conditions. These findings highlight its potential application in auditory rehabilitation and hearing technologies where ENV cues are critical. Full article

Background/Objectives: The Speech-Language Pathologists (SLP) are an integral part of the multidisciplinary team approach to rehabilitation of persons with aphasia (PWA). However, the efficacy of treatment provided by SLPs can vary due to several factors related to clinicians, patients, and the availability of services. The present study was conducted with the aim of investigating current practices in aphasia rehabilitation, key challenges, and future directions as perceived by the SLPs in the Indian context. Methods: The study was conducted using a web-based survey comprising a 32-item questionnaire to gather information related to demographic and professional details, knowledge and use of aphasia rehabilitation approaches, patient education, counselling, bilingual & multilingual contexts, and challenges faced by SLPs. A total of 142 responses were analyzed after initial screening to assess the knowledge, use, and confidence of aphasia rehabilitation along with challenges faced by SLPs in the Indian context. Results: The results indicated significant challenges in the assessment of aphasia due to a lack of formal screening and diagnostic languages in several languages. Further, the results also indicated variations in the knowledge level and confidence in the use of various approaches to aphasia rehabilitation, which warrants the urgent need for organizing short-term training programs for SLPs. The participants also self-reported significant challenges in managing bilingual and multilingual patients with aphasia due to differences in their knowledge and confidence in the selection of language for treatment. On the other side, major patient-related challenges include inadequate logistics, lack of funding, unavailability of speech and language therapy services, social acceptance, and support from family members. The participants also reported the necessity of improving tele-rehabilitation services and developing materials and mobile apps for rehabilitation in Indian languages as future directions for aphasia rehabilitation. Conclusions: The present study through a self-reported questionnaire identified key challenges in aphasia rehabilitation related to the clinician and PWA in the Indian context. The results of the study warrant the need for immediate action to overcome the challenges to enhance the rehabilitation services to PWAs. Full article

Objectives: This study aimed to evaluate auditory cortical maturation in pediatric cochlear implant (CI) users using speech-evoked cortical auditory evoked potentials (CAEPs) and to compare P1 latency responses with age-matched normal-hearing (NH) peers. Secondary objectives included examining the relationship between P1 latency, age, and duration of implant use to assess experience-dependent cortical plasticity. Materials and Methods: Seventy children were enrolled, including 40 prelingually deaf CI users and 30 NH controls matched for age and sex. CAEPs were recorded using the HEARLab system with three speech tokens representing low (/m/), mid (/g/), and high (/t/) frequencies, presented at 55 dB SPL in a free-field setup. The P1 component was identified as the first positive deflection between 50 and 150 ms after stimulus onset. Group comparisons were performed using Student’s t-test, and correlations between P1 latency, age, and implant-use duration were analyzed using the Pearson correlation test (p < 0.05). Results: Mean P1 latencies were significantly longer in CI users than in NH peers for the /m/ and /t/ stimuli (p = 0.036 and p = 0.045, respectively), while no significant difference was found for /g/ (p = 0.542). In NH children, P1 latency negatively correlated with age (r = −0.44, p < 0.05), indicating maturation-related shortening. Among CI users, longer implant-use duration was associated with shorter P1 latencies across all speech tokens (/m/: r = −0.37; /g/: r = −0.49; /t/: r = −0.43; p < 0.05 for all). Conclusions: Speech-evoked CAEPs provide a sensitive and objective measure of auditory cortical development in children with cochlear implants. P1 latency reflects both chronological and hearing-age-related maturation, supporting its clinical use as a biomarker for cortical plasticity and rehabilitation progress in pediatric CI care. Full article

This study investigates the influence of toxic language in hate speech targeting immigrants, particularly through narrative formats like first-person X (Twitter) threads. Hate speech, defined as promotion of hatred based on personal or group characteristics, increasingly escalates on social media, impacting public attitudes and behaviors. While previous research has primarily focused on measuring the scope of hate speech through content analysis and computational methods, there has been limited attention to its effects on audiences. This study presents the results of an online experiment (N = 339) with a 2 × 2 between-subjects design that manipulates the presence of toxic language and message popularity. Results indicate that hate messages lacking toxic language promote greater identity fusion with the author of the message, which in turn increases the intention to share the message, reinforces negative attitudes toward immigrants, and increases support for harsh policies against irregular immigration. Moreover, non-toxic hate messages significantly enhance narrative transportation exclusively for individuals with conservative political views, thereby further increasing their intention to share the message. These findings highlight that subtler forms of hate speech can create strong audience connections with hostile perspectives, emphasizing the need for anti-hate campaigns to address both overt and subtle hate content. Full article

Background/Objectives: In mainstream educational settings, deaf and hard-of-hearing (DHH) students may have limited or no access to the spoken lectures and discussions that are central to the hearing majority classroom. Yet, engagement in these educational and social exchanges is fundamental to their learning and inclusion. Two primary visual accessibility solutions can support this need: real-time speech-to-text transcriptions (i.e., captioning) and high-quality sign language interpreting. Their combined use (or coexistence), however, raises concerns of competition between concurrent streams of visual information. This article examines the empirical evidence concerning the effectiveness of using both captioning and sign language simultaneously in educational settings. Specifically, it investigates whether this combined approach leads to better or worse content learning for DHH students, when compared to using either visual accessibility solution in isolation. Methods: A review of all English language studies in peer-reviewed journals until August 2025 was performed. Eligible studies used an experimental design to compare content learning when using sign language and captions together, versus using sign language or captions on their own. Databases Reviewed: EMBASE, PubMed/MEDLINE, and PsycInfo. Results: A total of four studies met the criteria for inclusion. This limited evidence is insufficient to decide on the coexistence of captioning and sign language. Yet, it underscores the potential of captions for content access in education for DHH, even when sign language is available. Conclusions: The present article reveals the lack of evidence in favor or against its coexistence with sign language. With the aim to be constructive for future research, the discussion offers considerations on the attentional demands of simultaneous visual accessibility resources, the diversity of DHH learners, and the impact of current and forthcoming technological advancements. Full article