Deepgram Speech to Text
Deepgram Speech to Text turns recorded audio files into text with Deepgram’s Nova models. The text comes out punctuated and formatted (numbers, times and dates as digits) and, if you don’t set the language, Deepgram detects it. It returns one Data per file, with the transcript in the text field.
You need a Deepgram API credential with your Deepgram account key. See Credentials. The audio is sent to Deepgram to be transcribed.
This is not the voice node for calls: to listen to a caller during a live call, the Voice Agent uses the Deepgram STT node. This component waits for the whole file.
When to use it
- Transcribe long call or meeting recordings: it accepts very large files.
- Turn voice notes into text when you don’t know their language in advance.
- Transcribe medical consultations with specialised vocabulary, with the
nova-3-medicalmodel. - Let an agent transcribe audio when it needs to, using the component as a tool.
- Transcribe a list of audio files in one go with Batch Mode.
Configuration
| Field | What it does |
|---|---|
| API Key (required) | Your Deepgram key. The key is filled in for you when you choose your Deepgram credential. |
| Audio (required) | The audio (or video) file to transcribe: WAV, MP3, OGG, M4A, WEBM… Connect it to the component that brings the file, for example a webhook, a chat with attachments or storage. It accepts several files. |
| Language | Optional. The language of the audio, for example es or en. Type multi when the audio mixes languages. Empty by default: Deepgram detects the main language. |
Advanced options
| Field | What it does |
|---|---|
| Model | The Deepgram model. Default: nova-3, the most accurate. nova-3-medical is meant for medical vocabulary. nova-2, nova-2-phonecall and nova-2-meeting are also available; the last two work in English only. |
Output
The Data output returns one record per file:
{ "text": "Hola, buenos días. Su cita está confirmada para el martes a las 10 y media.", "language": "es", "file_name": "call.mp3", "provider": "Deepgram", "model": "nova-3", "duration": 5.1}duration is the length of the audio in seconds. With Language set to multi, language stays empty.
Each file can be up to 2 GB.
Example
A customer service team records its calls. Every night, the flow picks up the day’s recordings from storage, passes them to Deepgram Speech to Text with Batch Mode, and an agent summarises each transcript and flags the ones asking for a refund.
Related components
- Speech to text and text to speech: every component that transcribes audio or produces it.
- Deepgram Text to Speech: the opposite, turns text into audio with the same credential.
- Groq Whisper Speech to Text, Gemini Speech to Text and Whisper Local: other ways to transcribe audio.
Tips
- Deepgram writes numbers and times as digits: “a las diez y media” comes out as “a las 10 y media” or “a las 10:30”. Keep it in mind if another component looks for specific words in the text.
- Measured in our test environment on 3 November 2026: the sentence “Hola, buenos días. Su cita está confirmada para el martes a las diez y media.” was transcribed in 0.4 s, word for word.