# Transcription and captures (https://voicelabs.now/docs/transcription)



VoiceLabs can turn audio into text using its hosted **Whisper** speech-to-text, running on VoiceLabs' GPU servers.

## Captures [#captures]

Every transcription creates a **capture** on your account — a record of the audio (dictations, recordings, uploads) together with its transcript. Your recent captures, with transcripts, can be listed from the surfaces below.

## How to transcribe today [#how-to-transcribe-today]

Transcription is live on two surfaces:

* **MCP** — the `transcribe` tool accepts an audio file either as a public https URL (`audioUrl`, up to 10 MiB) or inline as base64 (`audioBase64`, up to 10 MiB of base64), with an optional language, and returns the transcript plus the created capture id. A payload that is not real audio (under 1 KiB, a bare header, or placeholder bytes) is refused without being transcribed. See [Connecting AI assistants](/connecting-ai-assistants).
* **API** — `POST /v1/transcriptions` transcribes a base64-encoded clip over REST with an API key, and `GET /v1/captures` lists your captures (paginated). See [Developer API](/api).

<Callout type="info" title="Metered server-side">
  Transcription runs on VoiceLabs' GPUs and is metered server-side by your plan. See [Usage and limits](/usage-and-limits).
</Callout>