Deepgram — AI Voice

Deepgram is an enterprise-grade Voice AI platform that provides real-time speech-to-text (STT), text-to-speech (TTS), and voice agent APIs. It solves the problem of slow, inaccurate, or costly transcription by using end-to-end deep learning models that process audio in under 300 milliseconds—without requiring pre-training on your specific audio.

What it does

Deepgram offers a unified API that converts audio to text (with streaming and batch options), generates natural-sounding speech, and orchestrates voice agents with built-in turn detection and interruption handling. It supports 45+ languages, speaker diarization, custom vocabulary, and automatic punctuation. Models like Nova-3 handle background noise, crosstalk, and far-field audio out of the box.

Who it's for

This API is designed for developers building voice-enabled apps (e.g., voice assistants, call analytics, live captioning), contact centers needing real-time call transcription, and media companies transcribing podcasts or videos at scale. It's less suited for one-off manual transcription jobs where a human editor is preferred, or for extremely low-budget hobby projects that don't need sub-second latency.

Real use cases

• Real-time captioning for live events and webinars
• Automated call scoring and sentiment analysis in contact centers
• Voice agent chat for customer support using Flux models
• Batch transcription of recorded meetings, interviews, and video content

• Real-Time Streaming — Transcribe audio as it's spoken via WebSocket API with sub-300ms latency
• Batch Processing — Upload pre-recorded files for asynchronous transcription
• Custom Vocabulary — Add industry jargon, names, or acronyms to improve accuracy
• Speaker Diarization — Identifies who spoke when in multi-person audio
• Punctuation & Formatting — Automatic capitalization, commas, and periods for readable transcripts
• Language Support — 45+ languages including English, Spanish, Mandarin, and Arabic
• Voice Agent API — Single unified API for STT, TTS, and LLM orchestration with turn detection

Deepgram pricing: from $0.0048/min. Billing model: Freemium.

For comparison: the median starting price in AI Voice is $10/month, measured across 54 tools we track. See the full SaaS Pricing Index →

Free Tier

Includes $200 in free credits to get started. No credit card required. Access to all public models with limited concurrency (up to 50 REST API, up to 50 WSS for STT).

Pay As You Go

No minimums, no expiration. $0.0048/min for Nova-3 Monolingual (pre-recorded), $0.0065/min for Flux English (streaming). Higher concurrency limits: up to 150 WSS for STT.

Growth

Pre-paid annual credits (from $4K+/year) save up to 20% vs pay-as-you-go. Includes increased concurrency: up to 225 WSS for STT, up to 60 for TTS and Voice Agent API.

All plans come with community and Discord support; premium SLAs available on Growth and Enterprise. Contact sales for custom models and enterprise deployment.

Find the right tool for you with our AI advisor →