Robust Nasality Representation Learning for Cleft Palate-Related Velopharyngeal Dysfunction Screening in Real-World Settings

Abstract

Velopharyngeal dysfunction (VPD) is characterized by inadequate velopharyngeal closure during speech and often causes hypernasality and reduced intelligibility. Although speech-based machine learning models can perform well under standardized clinical recording conditions, their performance often drops in real-world settings because of domain shift caused by differences in devices, channels, noise, and room acoustics. To improve robustness, we propose a two-stage framework for VPD screening. First, a nasality-focused speech representation is learned by supervised contrastive pre-training on an auxiliary corpus with phoneme alignments, using oral-context versus nasal-context supervision. Second, the encoder is frozen and used with lightweight classifiers on 0.5-second speech chunks, whose probabilities are aggregated to produce recording-level decisions with a fixed threshold. On an in-domain clinical cohort of 82 subjects, the proposed method achieved perfect recording-level screening performance (macro-F1 = 1.000, accuracy = 1.000). On a separate out-of-domain set of 131 heterogeneous public Internet recordings, large pretrained speech representations degraded substantially, while MFCC was the strongest baseline (macro-F1 = 0.612, accuracy = 0.641). The proposed method achieved the best out-of-domain performance (macro-F1 = 0.679, accuracy = 0.695), improving on the strongest baseline under the same evaluation protocol. These results suggest that learning a nasality-focused representation before clinical classification can reduce sensitivity to recording artifacts and improve robustness for deployable speech-based VPD screening.

Figures (2)

• Figure 1: Overview of the proposed two-stage framework. Stage 1: nasality representation pre-training using supervised contrastive learning (SupCon) with positive pairs sampled from the same speaker, same vowel, and same auxiliary class (oral-context vs. nasal-context), while restricting contrastive comparisons to within-vowel pairs to reduce phonetic-content leakage. A Wav2Vec2.0 backbone with trainable layer fusion and a projection head outputs 256-dimensional $\ell_2$-normalized embeddings. Stage 2: robust VPD screening under domain shift (lab $\rightarrow$ wild) using the frozen encoder as a feature extractor on 0.5 s chunks, followed by a lightweight classifier (LR/SVM/MLP/XGBoost) and mean aggregation of chunk-level probabilities to produce recording-level screening decisions (in-domain and out-of-domain) using a fixed decision threshold.
• Figure 2: UMAP visualization of SupCon nasality embeddings on the auxiliary validation split. Each panel shows a single vowel with a class-balanced subset of vowel-centered segments (0.20 s). Points are colored by the auxiliary nasality context label (oral_core vs. nasal_strong). Vowel labels follow ARPAbet notation from the forced-alignment annotations.