agenthub · audio.transcribe
Transcribe Audio
Turn an audio clip into text. The reverse of Voice Output, and usually the front door of a voice workflow.
Why the node earns its place
It receives a clip — typically a base64 audio URI from a browser microphone — and calls the transcription capability, so a downstream LLM or reply node can work in text like any other agent.
The dropdown lists transcription models only, for the same reason the other media nodes filter theirs: a model that cannot transcribe should not be selectable.
How it works at run time
- 1Take the clip from
audio, or fetchaudio_url. - 2Transcribe, passing the
languagehint when set; blank means auto-detect. - 3Return the transcript and usage.
What you wire
Inputs
- audiostring
- audio_urlstring
Outputs
- textstring
- usageobject
Configuration
| Key | Type | Default & options | What it does |
|---|---|---|---|
| model * | string | choices from model-library:transcription_models | Speech-to-text model to transcribe with |
| language | string | — | Optional ISO language hint (e.g. 'en'). Blank = auto-detect. |
What usually goes wrong
Watch for this
Set language when you know it. Auto-detect is good but not free of surprises on short clips, code-switching, or heavily accented speech — and a wrong detection produces confident nonsense rather than an error.
Behaviour & provenance