Deepgram

Production-grade speech-to-text, text-to-speech, and voice-agent APIs for real-time and batch audio.

Freemium· Free credits on signup; usage-based pricing; enterprise contracts availableAudioNova, Flux, Speak (proprietary)

Visit website →

Best for

Pick Deepgram if you're building a real-time voice product — agents, call analytics, live captions — and need streaming latency plus a self-hosting escape hatch.

Skip if

Skip it if you only need occasional batch transcription and would prefer a fully open-source stack like Whisper or a cheaper pay-per-minute API.

Deepgram is a voice AI platform built around a family of proprietary speech models: Nova for transcription, Flux for multilingual conversational STT across 10 languages, and Speak for text-to-speech. It exposes everything as low-latency APIs with both cloud and self-hosted deployment, plus a Voice Agent API that bundles STT, TTS, and LLM orchestration into a single conversational pipeline.

The target user is a developer or platform team building voice into a product — contact centers, meeting bots, medical scribes, podcast transcription, IVR replacements. Deepgram's differentiator against the OpenAI/Google/AssemblyAI pack has historically been latency and cost per hour at scale, plus the option to run models on-prem for compliance-heavy workloads. Pricing isn't posted on the marketing page, but the console offers self-serve signup with free credits, and metered usage scales into enterprise contracts.

It is not open source, and while docs are thorough, choosing between Nova, Flux, and the Agent API can be confusing at first. For teams that just want a hosted transcription call and don't need real-time streaming, cheaper batch-only alternatives exist. But if latency, streaming, or self-hosting matter, Deepgram is one of the most credible options.

Editor's take

Deepgram is one of the few voice AI vendors that consistently ships models competitive with the hyperscalers on latency and price. The Voice Agent API is a smart bet as the interaction pattern shifts from transcription to full conversations. Just budget time for a proper bake-off — Nova vs Whisper vs AssemblyAI is a real decision.

— The AI Tool Bible editorial team