做语音交互或角色扮演应用的开发者，终于有了一个能理解语气和情绪的端到端模型，建议直接试 API。
上海 AI 实验室 StepFun 于 2026 年 5 月发布 StepAudio 2.5 Realtime，这是一款端到端的实时语音大模型，支持中英文，通过 WebSocket API 连接。该模型在 2026 年 4 月的五项基准测试中均排名第一，包括 80.41 的人类评估分数和 82.18 的副语言理解分数。其特色在于角色扮演特定的 RLHF 训练和副语言理解能力，允许用户自定义角色风格。这标志着语音 AI 在情感和角色模拟方面取得了重要进展。
StepFun Releases StepAudio 2.5 Realtime: An End-to-End Voice Model with Roleplay-Specific RLHF and Paralinguistic Comprehension
StepFun, the Shanghai-based AI lab, released StepAudio 2.5 Realtime in May 2026 — an end-to-end real-time speech large language model with fully customizable persona capabilities. The model connects via a WebSocket API, supports Chinese and English, and ranked first across all five benchmark dimensions tested in April 2026, including an 80.41 human evaluation score and 82.18 on paralinguistic comprehension. The post StepFun Releases StepAudio 2.5 Realtime: An End-to-End Voice Model with Roleplay-Specific RLHF and Paralinguistic Comprehension appeared first on MarkTechPost .