Deepgram Language Detection: Supported Models and How to Use Them

Need to auto-detect a caller’s language before routing transcripts or running downstream NLP? Deepgram does support language detection, but only on specific models and only for pre-recorded audio. Here’s the reference I wish I had when wiring voice pipelines into n8n and Render.

Quick capabilities matrix

Model family	Example model IDs	Supports `detect_language`?	Streaming support?	Notes
Nova-3	`nova-3`, `nova-3-general`, `nova-3-medical`	✅ Yes (pre-recorded only)	❌ No	Flag `detect_language=true`. Streaming endpoints ignore it.
Nova-2	`nova-2`, `nova-2-general`, `nova-2-meeting`, `nova-2-phonecall`, etc.	✅ Yes (pre-recorded only)	❌ No	Works across domain-specific variants.
Nova (legacy)	`nova`, `nova-general`, `nova-phonecall`, `nova-medical`	✅ Yes (pre-recorded only)	❌ No	Still supported for batch jobs.
Enhanced	`enhanced`, `enhanced-general`, `enhanced-meeting`, `enhanced-phonecall`, etc.	✅ Yes (pre-recorded only)	❌ No	Use when you need higher accuracy but don’t require streaming.
Base	`base`, `base-general`, `base-meeting`, `base-phonecall`, etc.	✅ Yes (pre-recorded only)	❌ No	Entry-level models still handle detection for offline audio.
Whisper via Deepgram	`whisper-base`, `whisper-large`, etc.	❌ No	❌ No	Whisper uses its own detection. Deepgram’s `detect_language` flag is ignored and you can’t restrict language sets.

Important constraints

Language detection only works for batch (pre-recorded) requests. Streaming (/listen/stream) will not honor detect_language.
The detectable language set is smaller than the full transcription language list. Check Deepgram’s Language Detection docs if you need a particular locale.
Whisper models ignore the detect_language flag entirely; you can’t narrow or enforce languages when calling them through Deepgram.

Batch transcription example

curl --request POST \
  --header "Authorization: Token $DEEPGRAM_API_KEY" \
  --header 'Content-Type: audio/wav' \
  --data-binary @youraudio.wav \
  --url 'https://api.deepgram.com/v1/listen?model=nova-3-general&detect_language=true'

Swap nova-3-general for any of the supported models in the table above. Deepgram returns a detected_language field in the response, allowing you to branch workflows on the fly.

Production checklist

Batch-only: Ship your audio file and wait for the JSON response. Streaming integrations require your own language detection fallback.
Model pick: Choose a Nova 3/2, Enhanced, or Base variant if detection matters. Avoid Whisper unless you’ve already handled language outside of Deepgram.
Management API: Need to confirm what’s available to your account? Call GET /v1/models or GET /v1/projects/{project_id}/models for a live list before hard-coding IDs.

Available model names (September 2025)

Use any of the following in the model query parameter:

Nova 3: nova-3, nova-3-general, nova-3-medical
Nova 2: nova-2, nova-2-general, nova-2-meeting, nova-2-phonecall, nova-2-finance, nova-2-conversationalai, nova-2-voicemail, nova-2-video, nova-2-medical, nova-2-drivethru, nova-2-automotive, nova-2-atc
Nova (legacy): nova, nova-general, nova-phonecall, nova-medical
Enhanced: enhanced, enhanced-general, enhanced-meeting, enhanced-phonecall, enhanced-finance
Base: base, base-general, base-meeting, base-phonecall, base-finance, base-conversationalai, base-voicemail, base-video
Whisper (no detection): whisper-tiny, whisper-base, whisper-small, whisper-medium, whisper-large

When in doubt, query the Management API to confirm active names. Deepgram occasionally retires or renames variants.

Takeaways

Detection + streaming isn’t a thing. Plan for batch if you need auto-language pivoting.
Stick to Nova/Enhanced/Base for language detection. Whisper is transcription-only in this context.
Parameterize everything. Keep model and detect_language in config so you can toggle per project.

Pair this guide with your n8n automations or Render services, and you’ll know exactly which Deepgram knob to turn before launching multi-lingual voice features.