A5046198404- Language Development and Disorders
- Phonetics and Phonology Research
- Voice and Speech Disorders
- Neurobiology of Language and Bilingualism
- Speech Recognition and Synthesis
- Reading and Literacy Development
- Hearing Loss and Rehabilitation
- Assistive Technology in Communication and Mobility
- Human Pose and Action Recognition
- Forecasting Techniques and Applications
- Artificial Intelligence in Healthcare and Education
- Sensory Analysis and Statistical Methods
- Stuttering Research and Treatment
- Genetics and Neurodevelopmental Disorders
- Hearing Impairment and Communication
- Autism Spectrum Disorder Research
- Speech and Audio Processing
- Context-Aware Activity Recognition Systems
- Video Surveillance and Tracking Methods
- Linguistic Variation and Morphology
- Animal Vocal Communication and Behavior
- Family and Disability Support Research
University of Maryland, College Park
2024-2025
Google (United States)
2021-2024
Syracuse University
2020-2024
Haskins Laboratories
2023
New York University
2023
Memorial University of Newfoundland
2023
Montclair State University
2023
Recent empirical studies have highlighted the large degree of analytic flexibility in data analysis that can lead to substantially different conclusions based on same set. Thus, researchers expressed their concerns these researcher degrees freedom might facilitate bias and claims do not stand test time. Even greater is be expected fields which primary lend themselves a variety possible operationalizations. The multidimensional, temporally extended nature speech constitutes an ideal testing...
Abstract Background Residual speech sound disorder (RSSD) is a high-prevalence condition that negatively impacts social and academic participation. Telepractice service delivery has the potential to expand access technology-enhanced intervention methods can help remediate RSSD, but it not known whether remote associated with reduction in efficacy of these methods. This project will systematically measure outcomes visual-acoustic biofeedback when delivered in-person or online. Methods/design...
Publicly available speech corpora facilitate reproducible research by providing open-access data for participants who have consented/assented to sharing among different teams. Such can also support clinical education, including perceptual training and in the use of analysis tools.
This study examines how ultrasound biofeedback and intensive treatment distribution affect speech sound generalization during an evidence-based treatment, Speech Motor Chaining, for children with persisting errors associated childhood apraxia of (CAS).In a 2 × factorial randomized controlled trial, ages 9-17 years meeting CAS criteria were to receive (a) distributed (20 sessions twice weekly over 10 weeks) or hr in 5 weeks, Week 1) (b) without biofeedback. Due the COVID pandemic, some...
Purpose Prior studies report conflicting descriptions of the relationships between phonological awareness (PA), vocabulary, and speech perception in preschoolers with disorders. This study sought to determine nature these a sample school-aged children residual sound errors affecting /ɹ/. Method Participants included 110 aged 7;0–17;4 (years;months) impacting Data on perceptual acuity bias an /ɹ/ identification task, receptive PA were obtained. A theoretically empirically motivated path model...
Purpose Research comparing different biofeedback types could lead to individualized treatments for those with residual speech errors. This study examines within-treatment response ultrasound and visual-acoustic biofeedback, as well generalization untrained words, errors affecting the American English rhotic /ɹ/. We investigated whether some children demonstrated greater improvement in /ɹ/ during or biofeedback. Each participant received both types. Individual predictors of treatment (i.e.,...
Because lab accuracy of clinical speech technology systems may be overoptimistic, validation is vital to demonstrate system reproducibility-in this case, the ability PERCEPT-R Classifier predict clinician judgment American English /ɹ/ during ChainingAI motor-based sound disorder intervention.All five participants experienced statistically-significant improvement in untreated words following 10 sessions combined human-ChainingAI treatment.These gains, despite a wide range PERCEPThuman and...
To evaluate whether features of childhood apraxia speech identified in previous literature could be replicated a sample school-age children.A review was conducted to identify candidate that have been previously considered when differentiating from other types sound disorders. The recoverable blinded transcriptions multisyllable word repetitions (MSWR) were applied cohort 61 children, aged 7-17, classified as having (n=21) or non-CAS Speech Sound Disorder (SSD, n=40).One hundred and...
The effects of different acoustic representations and normalizations were compared for classifiers predicting perception children's rhotic versus derhotic /ɹ/. Formant Mel frequency cepstral coefficient (MFCC) 350 speakers z-standardized, either relative to values in the same utterance or age-and-sex data typical Statistical modeling indicated normalization significantly increased classifier performances. Clinically interpretable formants performed similarly MFCCs endorsed deep neural...
Mispronunciation detection tools could increase treatment access for speech sound disorders impacting, e.g., /ɹ/.We show age-and-sex normalized formant estimation outperforms cepstral representation of fully rhotic vs. derhotic /ɹ/ in the PERCEPT-R Corpus.Gated recurrent neural networks trained on this feature set achieve a mean test participantspecific F1-score =.81 (σx=.10,med = .83,n 48), with post hoc modeling showing no significant effect child age or sex.
Purpose To assess the concurrent validity of two tasks used to inform diagnosis childhood apraxia speech (CAS), this study evaluated agreement between Syllable Repetition Task (SRT) and Maximum Rate Trisyllables (MRR-Tri). Method A retrospective analysis was conducted with 80 children 7-16 years age who were referred for treatment studies. All had a sound disorder, all completed both SRT MRR-Tri. On each task, classified as meeting or not tool's threshold CAS based on sequencing errors...
This feasibility trial describes changes in rhotic production residual speech sound disorder following ten 40-min sessions including artificial intelligence (AI)-assisted motor-based intervention with ChainingAI, a version of Speech Motor Chaining that predicts clinician perceptual judgment using the PERCEPT-R Classifier (Perceptual Error Rating for Clinical Evaluation Phonetic Targets). The primary purpose is to evaluate /ɹ/ productions directly after practice ChainingAI versus before and...
Childhood apraxia of speech is a genetically driven, neurodevelopmental sound disorder with deficits theorized to reflect difficulty in the spatiotemporal programming movements. Therefore, this work examined how well articulatory coordination features generated from audio-estimated kinematic data distinguished speakers childhood versus non-apraxic disorder. Two correlation-based feature sets motivated by recent literature demonstrated high performance replicated 6-fold nested cross validated...
Abstract Purpose Typically developing children assigned male at birth (AMAB) and female (AFAB) produce the fricative /s/ differently: AFAB with a higher spectral peak frequency. This study examined whether implicit knowledge of these differences affects speech‐language pathologists’/speech language therapists’ (SLPs’/SLTs’) ratings accuracy, by comparing made in conditions where SLPs/SLTs were blind to children's sex (SAB) which they told this information. Methods SLPs ( n = 95) varying...
The advancement in deep learning and internet-of-things have led to diverse human sensing applications. However, distinct patterns sensing, influenced by various factors or contexts, challenge generic neural network model's performance due natural distribution shifts. To address this, personalization tailors models individual users. Yet most studies overlook intra-user heterogeneity across contexts sensory data, limiting generalizability. This limitation is especially critical clinical...
There has been a surge of interest in leveraging speech as marker health for wide spectrum conditions. The underlying premise is that any neurological, mental, or physical deficits impact production can be objectively assessed via automated analysis speech. Recent advances speech-based Artificial Intelligence (AI) models diagnosing and tracking mental health, cognitive, motor disorders often use supervised learning, similar to mainstream technologies like recognition verification. However,...
Purpose: This study evaluates the initial efficacy of Chaining SPeech Lessons in Intensive Ten-minute Sessions (SPLITS), an alternative service delivery model for Speech Motor treatment approach. We hypothesized that SPLITS would result improvements /ɹ/ accuracy on syllables and untrained words when compared to a no-treatment condition. Method: Within randomized controlled trial, thirteen 7–9-year-old children with difficulty producing were receive either immediately or after 8-week delay....
Background: Publicly-available speech corpora facilitate reproducible research by providing open-access data for participants who have consented/assented to sharing among different teams. Such can also support clinical education, including perceptual training and in the use of analysis tools.Purpose: In this Research Note, we introduce PERCEPT-R PERCEPT-GFTA corpora, which together contain over 36 hours audio (> 125,000 syllable, word, phrase utterances) from children, adolescents,...
Because lab accuracy of clinical speech technology systems may be overoptimistic, validation is vital to demonstrate system reproducibility - in this case, the ability PERCEPT-R Classifier predict clinician judgment American English /r/ during ChainingAI motor-based sound disorder intervention. All five participants experienced statistically-significant improvement untreated words following 10 sessions combined human-ChainingAI treatment. These gains, despite a wide range PERCEPT-human and...
We present the PERCEPT-R corpus, a labeled corpus of child speakers American English with typical speech and residual sound disorders affecting rhotics. demonstrate utility age-and-gender normalized formants extracted from in training support vector classifiers to predict ground-truth perceptual judgments “rhotic” (i.e., dialect-typical) clinical “derhotic” /ɹ/ for novel (mean participant-specific f-metrics = .83; SD .18, N 281).
Purpose: To assess the concurrent validity of two tasks used to inform diagnosis childhood apraxia speech (CAS), this study evaluated agreement between Syllable Repetition Task (SRT) and Maximum Rate Trisyllables (MRR-Tri).Method: A retrospective analysis was conducted with 80 children 7–16 years age who were referred for treatment studies. All had a sound disorder, all completed both SRT MRR-Tri. On each task, classified as meeting or not tool’s threshold CAS based on sequencing errors...