📄 中文摘要
实现低延迟的对话式 AI 一直是技术发展的目标，特别是实现“可打断的、低于500毫秒”的交互。研究表明，人类在对话中感知自然的时间间隔为200至500毫秒，超过800毫秒则会破坏交流的流畅性。为了应对这一挑战，过去的技术依赖于语音活动检测（VAD）和填充词等方法，但底层架构仍然是瓶颈。通过改进模型链的设计，尤其是语音转文本（STT）引擎的整合，有望提升语音交互的实时性和自然度。
与机器对话：使用 OpenAI 实时 API 构建低延迟语音代理
📄 English Summary
Talking to Machines: Building Low-Latency Voice Agents with OpenAI Realtime API
Achieving low-latency conversational AI has been a key goal in technology, particularly the aim for 'interruptible, sub-500ms' interactions. Research indicates that humans perceive a conversation as natural when the gap between speakers is between 200 and 500 milliseconds; anything longer than 800 milliseconds disrupts the flow of communication. To address this challenge, past technologies have relied on methods such as Voice Activity Detection (VAD) and filler words, but the underlying architecture remains a bottleneck. By improving the design of model chains, especially the integration of Speech-to-Text (STT) engines, it is possible to enhance the real-time nature and naturalness of voice interactions.