基於大型語言模型之合成普通話語音於植入人工耳蝸聽障成人之感知研究

// Perception of Synthesized Mandarin Speech Based on a Large-Scale Language Model Among Deaf Adults With Cochlear Implants

本研究比較大型語言模型（LLM）合成的普通話句子與自然語音對植入人工電子耳（CI）聽障成人的可懂度｜50名受試者（25名CI使用者、25名正常聽力）在不同語速下完成句子辨識測試｜結果顯示CI用戶辨識率較低、反應較慢，但當語速相同時，LLM合成語音與自然語音的可懂度無顯著差異｜研究指出LLM合成語音具備輔助語音感知與康復的潛力。

// This study compared the intelligibility of Mandarin sentences synthesized by a large-scale language model (LLM) to natural speech for deaf adults with cochlear implants (CIs). Fifty participants (25 with CIs, 25 normal hearing) completed sentence recognition tasks with natural and synthetic speech at different speaking rates. Results showed CI users had lower recognition and slower reactions overall, but no significant difference between natural and LLM-based synthetic speech when speaking rates were matched. The findings suggest LLM-based synthetic speech is as intelligible as natural speech for CI users and could aid speech perception and rehabilitation.

Published 2026年6月29日