Deepgram
Voice AI infrastructure: speech-to-text, text-to-speech, and a voice agent API
Last reviewed 2026-06-19
Deepgram is a voice AI infrastructure platform: a developer platform of real-time APIs backed by proprietary speech models. Its core offerings are speech-to-text (the Nova family), text-to-speech (Aura), audio intelligence (summarization, topics), and a Voice Agent API that combines STT, an LLM, and TTS over a single WebSocket with function calling and SDKs. It is not an autonomous agent itself; it provides the building blocks that developers and enterprises use to build voice agents. Founded in 2015 by Scott Stephenson (Y Combinator W16), Deepgram is based in San Francisco and serves contact centers, telephony, healthcare, and government, including air-gapped on-prem deployments. It raised a reported $130M Series C in January 2026 at a $1.3B valuation. Because it is infrastructure, its capabilities are assistant-level building blocks; "powering voice agents" is developer-controlled rather than autonomous.
What it can do
Transcribe speech (Nova)
AssistantReal-time and pre-recorded speech-to-text via the Nova model family across many languages.
sourceSynthesize speech (Aura)
AssistantText-to-speech via the Aura model family for natural-sounding voice output.
sourcePower voice agents via the Voice Agent API
SupervisedA single WebSocket API combines STT, an LLM, and TTS with function calling so developers can build voice agents; the agent logic is developer-controlled.
sourceAnalyze audio
AssistantAudio intelligence features such as summarization and topic detection over recorded audio.
source
Strengths
- +Mature, competitive STT (Nova) with low per-minute pricing and strong streaming
- +Rare true self-hosted, on-prem, and air-gapped options for regulated and government use
- +A single Voice Agent API collapses the STT-LLM-TTS stack
Limitations
- −Infrastructure, not a finished product: you build the agent and UX yourself
- −The Voice Agent API is materially pricier, and connection-time billing can surprise
- −Quality and cost vary by model and language tier
Overview
Deepgram is a voice AI infrastructure platform: real-time APIs backed by proprietary speech models. It provides the building blocks for voice agents rather than being an agent itself.
What it does
Its core offerings are speech-to-text (Nova), text-to-speech (Aura), audio intelligence (summarization, topics), and a Voice Agent API that combines STT, an LLM, and TTS over a single WebSocket with function calling and multi-language SDKs.
Integrations & setup
Works with Twilio, LiveKit, Vapi, and Amazon SageMaker, with SDKs in Python, JS, C#, and Go. Available as shared cloud, dedicated single-tenant, and self-hosted/on-prem/air-gapped.
Pricing
Pay-as-you-go with a free credit. Reported rates include Nova-3 streaming STT around $0.0048/min, Aura-2 TTS around $0.030/1k characters, and the Voice Agent API around $0.075/min, plus Growth and Enterprise tiers.
Best for / not for
Best for developers and enterprises building voice products, especially those needing on-prem or air-gapped speech. Not a fit for buyers wanting a turnkey agent.
Traction
Deepgram raised a reported $130M Series C in January 2026 at a $1.3B valuation led by AVP, with strategic investors including Twilio, ServiceNow Ventures, and SAP, bringing reported total funding above $215M. It states it serves 200,000+ developers.
Alternatives
Cartesia is the closest voice-model infrastructure competitor; ElevenLabs and Vapi overlap on voice agents.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. Deepgram isn't tracked yet, want it added? Request tracking.
FAQ
Is Deepgram an AI agent?+
No. Deepgram is voice AI infrastructure: speech-to-text, text-to-speech, audio intelligence, and a Voice Agent API. Its capabilities are building blocks developers use to build voice agents, so they are assistant-level rather than autonomous.
Can Deepgram run on-premise?+
Yes. It offers self-hosted, on-prem, and air-gapped deployments, which is a key reason regulated and government customers use it.
Sources
- Deepgram (official site) · accessed 2026-06-19
- Deepgram developer documentation · accessed 2026-06-19
- Deepgram pricing · accessed 2026-06-19
- Deepgram raises $130M Series C at $1.3B valuation (press release) · accessed 2026-06-19
Last reviewed 2026-06-19