Researchers have developed CASA, a new system for automatic speaking assessment that uses a combination of the Whisper-medium and Qwen3.5-2B large language models. CASA achieves state-of-the-art performance with improved interpretability by separating acoustic and content contributions to predictions. The system demonstrated a root mean square error of 0.358 on the Speak & Improve Corpus 2025, outperforming previous methods while using fewer inference parameters. CASA's architecture is designed for easy adaptation to other speaking assessment datasets and includes handcrafted fluency features. AI
IMPACT Introduces a more interpretable and efficient approach to automated speaking assessment using LLMs.
RANK_REASON The cluster contains an academic paper detailing a new model and methodology for speaking assessment. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
