Adapting PhonoQ-2.0 for Improved Child Speech Phonology Analysis
October 2, 2026
Adapting PhonoQ-2.0 to child-specific data improves voicing recognition macro-F1 from 0.922 to 0.987. The method also significantly improves manner recognition, achieving 0.804 macro-F1 compared to 0.70 when using models supervised only on adult speech.
HOW THIS AFFECTS YOU
●
researcherYou can use child-adapted alignment supervision to fix performance gaps in speech embeddings.
●
healthThis provides more accurate, interpretable phonological analysis for pediatric speech diagnostics.