AI Gateway Adds Support for Streaming Audio Transcription

AI Gateway now offers beta support for streaming transcription. Previously, you had to send an entire audio file and receive the completed transcription result all at once, but now you can stream audio as soon as it's captured and receive the transcription results generated by the model in real time.

This is available through the AI SDK's streamTranscribe function, which works by streaming raw PCM audio to the openai/gpt-realtime-whisper model and receiving partial transcription results via transcript-delta events. By simply swapping the model string, it works the same way with other streaming-supported models like xai/grok-stt.

It can also be used to implement voice-input-based agents. In this structure, user speech is streamed to a transcription model to be converted into real-time text, which is then passed to the agent. Since the agent itself receives plain text, this can be applied directly to existing text-based agents.