Paediatric-HGNN: A Hybrid Heterogeneous Graph Neural Network for Detecting Disfluency in Children's Speech via Multiscale Acoustic Fusion
Abstract
Automated stuttering detection (ASD) systems struggle with paediatric speech due to high acoustic variability in developing voices and the subtle distinction between pathological stuttering and typical developmental disfluencies. We introduce Paediatric-HGNN, a framework using a Context-aware Part-whole Interaction Network (CaPIN) tailored for paediatric data. Instead of conventional 1D signal modelling, our approach builds a heterogeneous graph capturing hierarchical relationships between lexical units (word nodes) and fine-grained acoustic segments (frame nodes). Trained on curated paediatric corpora (UCLASS and FluencyBank), Paediatric-HGNN achieves 82.4% weighted accuracy and a Typical Disfluency F1-score of 0.386. Modelling hierarchical lexical-acoustic interactions captures developmental "searching" behaviour, offering a more robust and interpretable tool for early clinical intervention.
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
The paper presents Paediatric-HGNN, a novel framework for detecting disfluency in children's speech, leveraging a heterogeneous graph neural network approach to capture complex interactions between lexical and acoustic features. This work significantly advances the field of automated speech disorder detection by addressing the unique challenges posed by pediatric speech variability and providing a clinically interpretable solution.
Methodology
The methodology presented in the paper is robust and innovative, utilizing a heterogeneous graph neural network (GNN) framework specifically designed for pediatric speech analysis. The introduction of the Context-aware Part-whole Interaction Network (CaPIN) is a significant advancement over traditional 1D signal modeling, allowing for the capture of complex hierarchical relationships between lexical and acoustic features. The use of a hybrid feature vector that combines self-supervised embeddings with traditional acoustic features demonstrates a thoughtful integration of modern techniques with established methods. The hierarchical attention mechanism enhances interpretability, which is crucial for clinical applications. However, the paper could benefit from a more detailed explanation of the graph construction process and the implications of the chosen feature dimensions.
Results & evaluation
The experimental evaluation is thorough, employing a well-structured 5-fold cross-validation protocol that ensures the model's robustness and generalizability. The performance metrics, including weighted accuracy and F1 scores, are appropriately chosen and reported with mean and standard deviation, providing clear insights into the model's effectiveness. The comparison against state-of-the-art models and the exploration of transfer learning from adult datasets highlight the challenges of domain adaptation in pediatric speech. The results indicate that the proposed model significantly outperforms existing methods, particularly in distinguishing between typical and core stuttering, which is a critical clinical need.
Reproducibility
The paper provides a comprehensive description of the datasets used, the feature extraction process, and the model architecture, which facilitates reproducibility. However, the absence of a publicly available code repository or demo URL limits the ability for other researchers to replicate the findings directly. Including such resources would greatly enhance the paper's impact and usability within the research community.
Limitations
One limitation of the study is the reliance on a relatively small dataset of pediatric speech, which may affect the model's generalizability to broader populations. Additionally, while the model shows promise in distinguishing between disfluency types, the lower performance in detecting core stuttering suggests that further refinement and larger datasets may be necessary to improve accuracy in this area. The paper also does not address potential biases in the dataset or the implications of using a specific demographic of children.
Broader impact
The implications of this research are significant, as it addresses a critical gap in automated stuttering detection for pediatric populations. The ability to accurately differentiate between typical developmental disfluencies and pathological stuttering can lead to earlier and more effective interventions, potentially improving communication outcomes for children. The interpretability of the model further enhances its clinical utility, making it a valuable tool for speech-language pathologists. The paper presents Paediatric-HGNN, a novel framework for detecting disfluency in children's speech, leveraging a heterogeneous graph neural network approach to capture complex interactions between lexical and acoustic features. This work significantly advances the field of automated speech disorder detection by addressing the unique challenges posed by pediatric speech variability and providing a clinically interpretable solution.
Author of this paper? Send a correction.