Automated Pronunciation Evaluation for Korean Toddler Speech using Speech Diarization and Self-Supervised Learning

Researchers have made advancements in automated pronunciation evaluation for Korean toddler speech and visual representation learning, achieving state-of-the-art performance in several benchmarks.

A study published on arXiv^[1] presents an end-to-end pipeline for automated pronunciation evaluation of Korean toddler speech, addressing a significant gap in assessment tools for Korean pediatric communication disorders, which affect approximately 44% of cases^[1]. The researchers created a novel corpus of 53 recordings from Korean-speaking children aged 2-5 years. Using NeMo SortFormer for speaker diarization, they achieved 88.69% speaker count accuracy and 33.04% diarization error rate (DER)^[1]. For pronunciation scoring, a cross-model ensemble achieved balanced accuracies of 0.720 for consonants and 0.845 for vowels, with a mean of 0.782. Meanwhile, another study on V-JEPA 2.1 reported state-of-the-art performance on several challenging benchmarks, including a 20-point improvement in real-robot grasping success rate over V-JEPA-2 AC^[2]. The model also demonstrated strong performance in robotic navigation, depth estimation, and global recognition, achieving 7.71 mAP on Ego4D for short-term object-interaction anticipation and 40.8 Recall@5 on EPIC-KITCHENS for high-level action anticipation^[2].