OpenAI推出GPT-Realtime-2语音模型支持70种输入语言实时翻译
OpenAI于周四宣布其API新增多项语音智能功能,包括GPT-Realtime-2、GPT-Realtime-Translate和GPT-Realtime-Whisper。其中GPT-Realtime-2基于GPT-5级推理能力构建,可处理复杂用户请求并实现拟真对话交互。
新模型支持70种输入语言和13种输出语言的实时翻译,翻译过程与对话节奏同步。GPT-Realtime-Whisper提供实时语音转文本功能,适用于会议记录和无障碍场景。系统内置内容安全机制,可自动中断违反有害内容指南的对话。
此次更新推动语音接口从基础应答向具备推理与执行能力的方向发展,适用于客服、教育及媒体等多领域。
GPT-Realtime-2采用GPT-5级推理架构
支持70种语言输入13种输出实时翻译
新增实时语音转文本与安全中断机制
Title:
OpenAI Launches GPT-Realtime-2 70 Input Languages Real-time Translation Transcription
Summary:
OpenAI has introduced new voice intelligence features in its API, including GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. The GPT-Realtime-2 model leverages GPT-5-class reasoning to handle complex user requests, enabling more natural and responsive voice interactions. GPT-Realtime-Translate supports real-time conversation translation across 70 input and 13 output languages.
The new transcription tool, GPT-Realtime-Whisper, delivers live speech-to-text conversion during interactions, enhancing accessibility and usability. These models collectively shift voice interfaces from basic call-and-response systems to dynamic tools capable of listening, reasoning, translating, and acting in real time. OpenAI emphasizes that the updates are designed for enterprise applications, particularly in customer service.
The features target sectors such as education, media, events, and creator platforms, expanding AI’s role in multilingual and interactive environments. OpenAI has implemented content moderation guardrails to prevent misuse, including automated conversation halts for policy violations. This release reflects a broader industry trend toward real-time, multimodal AI integration.
Key Takeaways:
GPT-Realtime-2 uses GPT-5-class reasoning for complex voice interactions
GPT-Realtime-Translate supports 70 input and 13 output languages
Live transcription via GPT-Realtime-Whisper enables real-time speech-to-text
Guardrails prevent misuse in spam, fraud, and harmful content generation
Source: Original Article
查看原文 →
View Original →