Vercel added two ways to transcribe audio through AI Gateway. An application can submit a finished recording in one request, or use Gemini 3.5 Transcribe Live to receive text while audio continues arriving. The announcement makes the implementation difference concrete: live transcription uses a WebSocket connection that stays open for incoming audio and returning text.
Its live example feeds 16 kHz, 16-bit PCM audio chunks into an experimental streamingSending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.Full definition → function in AI SDK V7. That provides a connection for microphone-driven applications, while the complete-recording function suits files already on disk. Developers need to choose the interface that matches when their application receives the audio.
Terms in this piece · Glossary
streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
