Build Live Translation Apps with GPT-realtime-translate (developers.openai.com)

OpenAI has announced the gpt-realtime-translate model, a specialized live speech-to-speech translation system designed for multilingual interactions in broadcasts, streams, calls, and video conversations. This model stands out for its optimization in interpretation, having been trained on thousands of hours of professional interpreter audio. It processes spoken language while simultaneously streaming translated audio, achieving low latency and high accuracy—key requirements for natural live interpretation. Unlike general-purpose voice models, gpt-realtime-translate focuses solely on translation, avoiding unnecessary interactions that can disrupt fluent communication. This innovative model allows developers to create applications for diverse use cases, including broadcast-style translations for webinars and conference keynotes, as well as conversational translations for call centers and video chats. The platform supports over 70 input languages and 13 output languages, with features such as dynamic voice adaptation that captures the tone and style of the original speaker. Through practical demos, developers can integrate live translation into existing audio paths via browser tabs, phone calls, and video calls, further enhancing accessibility and communication across language barriers in real-time scenarios.