AI Gateway now supports streaming transcription
By Steven Van ·
AI Gateway's new streamTranscribe function returns live transcript deltas for models like openai/gpt-realtime-whisper and xai/grok-stt.
Vercel's AI Gateway can now stream transcription results as audio is captured, instead of waiting for a complete file and returning the full transcript in one response. The feature, in beta, is available through the AI SDK's streamTranscribe function with any streaming-capable transcription model, such as openai/gpt-realtime-whisper or xai/grok-stt, and returns transcript deltas as well as partial and final transcripts.
Vercel points to live captioning and voice input as the main uses, and notes it can add a voice mode to an existing text-based agent: speech is streamed to a transcription model and the resulting text passed straight to the agent, with no change needed on the agent side. Full details are in the streaming transcription documentation.