Deep Supervised Contrastive Learning of Pitch Contours for Robust Pitch Accent Classification in Seoul Korean

2026-04-21 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

Dual-Glob, a deep supervised contrastive learning framework, has been developed to classify fine-grained pitch accent patterns in Seoul Korean. This framework addresses the challenge of mapping continuous F0 contours to discrete tonal categories within the Autosegmental-Metrical (AM) model of intonational phonology, which is complicated by variable F0 realizations in natural speech. Dual-Glob captures holistic F0 contour shapes by ensuring structural consistency between clean and augmented views in a shared latent space, departing from conventional local predictive models. The researchers also introduced the first large-scale benchmark dataset for this task, comprising 10,093 manually annotated Accentual Phrases in Seoul Korean. Experimental results demonstrate that Dual-Glob achieves a state-of-the-art accuracy of 77.75% and an F1-score of 51.54%, significantly outperforming strong baseline models.

Key takeaway

For research scientists working on intonational phonology or speech processing for tonal languages, Dual-Glob offers a robust methodology for classifying pitch accent patterns. You should consider integrating deep supervised contrastive learning to capture holistic F0 contour shapes, especially when dealing with the variability of real-world speech. This approach can significantly improve classification accuracy and F1-score compared to traditional local predictive models.

Key insights

Dual-Glob uses deep supervised contrastive learning to robustly classify Seoul Korean pitch accent patterns from continuous F0 contours.

Principles

• Holistic F0 contour shapes are crucial.
• Structural consistency improves classification.
• Data-driven methods support AM phonology.

Method

Dual-Glob employs deep supervised contrastive learning to enforce structural consistency between clean and augmented F0 contour views in a shared latent space, enabling robust classification of pitch accent patterns.

In practice

• Use contrastive learning for F0 contour analysis.
• Develop large, annotated datasets for phonology.
• Apply holistic feature capture in speech tasks.

Topics

• Deep Supervised Contrastive Learning
• Pitch Accent Classification
• Seoul Korean Phonology
• F0 Contour Analysis
• Dual-Glob Framework

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Contrastive and Adversarial Disentanglement for Speaker Representations in Brazilian Portuguese — Combining adversarial and contrastive learning improves speaker representation disentanglement from environmental factors.
• A frequency analysis of filterbank initialisation and noise augmentation for LEAF — LEAF filterbanks show high inertia, resisting adaptation and converging to a fixed S-curve regardless of input frequency content.
• Incremental learning for audio classification with Hebbian Deep Neural Networks — Hebbian learning with kernel plasticity improves incremental audio classification by balancing new learning and knowledge retention.
• Joint-Centric Dual Contrastive Alignment with Structure-Preserving and Information-Balanced Regularization — HILBERT aligns audio-text representations using dual contrastive learning and regularization for long sequences in low-resource settings.
• PASQA: Pitch-Accent-Focused Speech Quality Assessment Model Trained on Synthetic Speech with Accent Errors — PASQA accurately assesses pitch-accent correctness in synthetic speech by leveraging a specialized dataset and architectural enhancements.
• Lung-SRAD: Spectral-Aware Regularized Audio DASS with Dual-Axis Patch-Mix Contrastive Learning for Respiratory Sound Classification — Lung-SRAD uses State Space Models and spectral-aware regularization to improve respiratory sound classification by preserving high-frequency details.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.