---
title: "Audio transcription"
description: "Audio transcription"
---

`audio_transcribe` turns a recording stored in Siglata, such as a WhatsApp voice note (.ogg/opus), MP3, M4A, WAV, FLAC or WebM, into text with speaker turns and timestamps. ElevenLabs, Siglata's transcription provider, does the transcription. Upload a local file first with `upload_begin`. Each started minute is charged to the organization's credits ($0.22 an hour at provider cost, or $0.27 with keyterms). Siglata reserves the cost before calling ElevenLabs. With too few credits it refuses with `credits_exhausted` before sending anything. If the provider cannot transcribe, the reservation is released. A failed charge write also returns `transcription_unavailable`, without reporting success. Recordings are capped at 30 minutes and 25 MiB (`audio_too_long`, `audio_too_large`). A file whose duration Siglata cannot read fails `audio_unsupported`.