Transcribe audio with Duaer Transcribe

Transcribe a recording

1. Search for Duaer Transcribe and connect it after a node that carries the audio file, such as a form upload or Read Files.
2. In Input Binary Field, enter the binary field that holds the recording. The default is data.
3. Open the Model dropdown and pick a transcription model.
4. Select Execute step. Each item keeps its fields and file and gains a transcript field with the text.

Duaer Transcribe accepts mp3, wav, m4a, ogg, flac, webm, and aac, up to 12 MB per file. When the format is wrong or the file is too large, Duaer Transcribe stops before calling the model and charges nothing.

Duaer Transcribe options

• Language: the spoken language as a code, such as en or zh. Leave it empty to let the model detect it; setting it when you know the language improves accuracy.
• Output Field: the field that holds the text. The default is transcript.

Pass the transcript to Duaer Judgment to classify intent or mood, or to Duaer Embedding for search.

Which models Duaer Transcribe offers

The Model dropdown lists every OpenRouter transcription model your Duaer admin enabled under Admin → Model pool, such as Voxtral Mini Transcribe, AssemblyAI Universal, or Fish Audio Transcribe. Speech recognition models on Volcengine Ark and Alibaba Bailian do not work with Duaer Transcribe yet and stay out of the list.

Prefer a dedicated model with Transcribe or STT in its name. General voice chat models sometimes reply to the recording instead of writing it down word for word.

Each recording is charged to the workflow owner's AI credits by its length. For credit allowances, refer to Duaer plans and credits.