Skip to main content
Deepgram’s speech AI platform — Nova-3 for pre-recorded transcription with audio intelligence (sentiment, topics, entities, intents, summaries), Flux for real-time streaming, and Aura-2 for text-to-speech — pairs with Mavera to turn spoken content into structured marketing intelligence. These five jobs chain Deepgram endpoints with Mavera surfaces to analyze meetings with full audio intelligence, mine podcasts for content strategy, extract messaging playbooks from customer calls, produce live blog posts from streaming event transcription, and build customer-authentic brand voices from interview recordings.

​
API Reference Card

DetailValue
Base URLhttps://api.deepgram.com
AuthToken — Authorization: Token dg_xxxxx
Key modelsNova-3 (pre-recorded), Flux (streaming), Aura-2 (TTS)
Key endpointsPOST /v1/listen (transcription), WSS /v1/listen (streaming), POST /v1/speak (TTS)
Audio intelligenceSummarization, topic detection, entity detection, sentiment analysis, intent detection, diarization
Mavera basehttps://app.mavera.io/api/v1
Mavera authAuthorization: Bearer mvra_live_...
All examples use two environment variables: DEEPGRAM_API_KEY (your Deepgram key starting with dg_) and MAVERA_API_KEY (your Mavera key starting with mvra_live_). Deepgram uses Authorization: Token — not Bearer. Never commit either key to version control.

​
Prerequisites

1

Deepgram API key

Sign up at console.deepgram.com. Create a new API key from the dashboard.
2

Mavera API key

Get your key from Mavera dashboard.
3

Install dependencies

# Python
pip install requests websockets

# JavaScript
npm install ws
4

Set environment variables

export DEEPGRAM_API_KEY="dg_your_deepgram_key"
export MAVERA_API_KEY="mvra_live_xxxxx"

​
Jobs

#JobDeepgram EndpointMavera SurfaceOutput
1Audio Intelligence → Meeting AnalysisPOST /v1/listen (Nova-3 + all intelligence)Mave Agent (Chat)Structured meeting intelligence
2Podcast Analysis PipelinePOST /v1/listen (topics, paragraphs, utterances)Mave Agent (Chat)Content strategy from podcast insights
3Customer Call IntelligencePOST /v1/listen (diarize, sentiment, utterances)Mave Agent (Chat)Messaging playbook from customer language
4Live Event Transcription → Real-Time ContentWSS /v1/listen (Flux streaming)GenerateLive blog posts from event stream
5Voice of Customer → Brand VoicePOST /v1/listen (batch interviews)Brand VoicesCustomer-authentic brand voice profile

​
Rate Limits & Production Notes

Deepgram FeatureLimitStrategy
Pre-recorded /v1/listen (Nova-3)Concurrency-based (plan-dependent)Queue files; 2s delay between calls
Streaming /v1/listen (Flux/Nova-3)1 connection per concurrent streamReconnect with exponential backoff
Audio intelligence featuresSame as base transcriptionNo extra limits — features are query params
File sizeUp to 2 GB per requestSplit larger files with ffmpeg
Deepgram charges by audio duration, not API calls. A 60-minute file costs the same as one request or six 10-minute chunks. Prefer single large requests. Monitor usage at console.deepgram.com.
Deepgram uses Authorization: Token — not Bearer. Every request needs Token {key}. The most common integration error is using Bearer, which returns 401. For streaming, pass the token as a WebSocket header — not a query parameter. Monitor Mavera credits at Dashboard.

All Integrations

Deepgram API Docs

Mave Agent

Generate

Brand Voice

Focus Groups