Integrating SAVI-Based AI Speech Recognition in L2 Chinese Story Retelling: Empirical Needs Analysis and Framework for Listening Comprehension

Abstract

Listening comprehension in Chinese as a Second Language (CSL) poses significant cognitive challenges for beginner learners, primarily driven by rapid acoustic streams and severe oral response anxiety. This study investigates the integration of an AI-assisted written story retelling framework grounded in the SAVI (Somatic, Auditory, Visual, Intellectual) accelerated learning model and structured through the ADDIE instructional design framework. A preliminary empirical needs analysis (N=78) revealed that while learners appreciate traditional instructional delivery, 75.6% struggle with continuous rapid speech streams and 60.3% suffer from speaking anxiety, highlighting the limitations of conventional audio-only drills and immediate oral tasks. To address these barriers, a four-phase SAVI-driven system architecture was developed: Phase 1 utilizes multi-speed native audio (89.7% preference) and dual-coding visual transcripts (80.8% preference); Phase 2 operationalizes somatic kinesthetic pathways via tactile character typing/writing (43.6% preference) to reduce affective filters; Phase 3 delivers real-time AI Natural Language Processing (NLP) semantic feedback (82.1% preference); and Phase 4 provides targeted micro-listening remediation by linking text gaps directly to source audio timestamps. By replacing high-anxiety oral execution with tactile written production, the proposed framework offers extended working memory processing time, promotes metacognitive gap identification, and establishes an active, self-regulated learning cycle. This study demonstrates that bridging Swain’s Output Hypothesis with multi-sensory AI scaffolding provides a scalable, innovative solution for enhancing L2 Chinese listening comprehension accuracy and mitigating classroom anxiety.