<span id="转写音频"></span>

# Transcribe audio

Select an available speech-recognition model, upload a short recording, and read its complete transcript. Use an audio workflow for batch processing and knowledge-base ingestion.

## Prerequisites

Prepare a credential with Genesis model permissions and your Genesis service address according to [Prepare to call an API](../../api-call-basics.md). [Find and select a model](models.md) that supports speech recognition and prepare a local audio file.

## Call sequence

1. [Find and select a model](models.md) for speech recognition and preserve its public model ID.
2. Prepare a short audio file (WAV, MP3, M4A, WebM, OGG, or FLAC) and upload it via [Transcribe audio](../../../../reference/api/genesis/audio-transcriptions.md) as multipart form data.
3. Read the complete synchronous transcript from the response and preserve the model identifier, request ID (if returned), and usage details.

## API operations

| Task | API operation | When to use it or what to check |
| --- | --- | --- |
| Upload short audio and obtain a transcript synchronously | [Transcribe audio](../../../../reference/api/genesis/audio-transcriptions.md) | Use for synchronous transcription of a single short audio recording. |

## Confirm the result

Verify that `text` in the response contains the expected transcript, and record `model`, `request_id` (if returned), and `usage`. Upstream success without valid text is returned as an error.

## Limitations and notes

Base64-encoded audio must not exceed 10 MiB (roughly 7.5 MiB raw) and must fit within your deployment's request-size limit. Synchronous Qwen3-ASR Flash is intended for recordings of about five minutes or less and provides no timestamps or speaker diarization. For batch parsing, see [Transcribe Audio in Bulk with a Workflow](../../../../tutorials/workflow-audio-transcription.md).
