Deepgram

Foundation API for speech.

Free planVoice Audio & Music

Visit deepgram.comCompare

+48.5%Search trend (3 mo) · DataForSEO

15,708Monthly searches

VoicePrimary category

N/ANo reviews yet

About Deepgram

If you've tried to build a voice agent, you know where it falls apart. You wire up a speech-to-text vendor, pipe transcripts into an LLM, send the reply to a separate text-to-speech service, and hope it all feels like a conversation. It usually doesn't. Every hop adds latency, the agent talks over the caller, it can't tell when someone has finished a sentence, and you end up debugging three vendors' quirks at once. Turn-taking and interruptions are where most voice bots sound robotic, and generic transcription models weren't built to handle them.

Deepgram's pitch is that it fixes this at the model layer. Flux STT is described as conversation-aware: it detects end-of-turn and interruptions and streams transcripts in real time. Flux TTS is pitched as a voice model that "thinks before it speaks," holding tone and continuity across a whole conversation, with latency as low as 80ms. On top of that, the Voice Agent API folds speech-to-text, LLM orchestration (context, prompt updates, response injections, function calling) and text-to-speech into a single API call. You bring your business logic and a transport layer, including telephony partners over PSTN or SIP. You can run it in real time or batch, in the cloud or self-hosted, which matters if you have compliance constraints. The customer list is a credible signal: Granola uses it for meeting transcription, and voice agent platforms like Vapi and Daily (the team behind the open source Pipecat framework) build on its APIs.

The free entry point is generous. You get $200 in credit with no credit card, no minimums and no expiration, and that tier allows up to 150 concurrent WebSocket streams for speech-to-text. The catch is that this is a developer product through and through. There's no app to click around in beyond a playground, support below Enterprise is community and Discord only, and some caps (5 concurrent streams for Deepgram Whisper Cloud, 10 for Audio Intelligence) are tight. If you're a developer or product team shipping voice features, it's one of the strongest starting points available. If you wanted a finished tool, it isn't one.

Key features

Flux STT (conversation-aware speech recognition)

Flux STT detects end-of-turn and interruptions while streaming transcripts in real time, so your agent knows when to speak and when to stop.

Flux TTS

Deepgram's newest text-to-speech model is built for live conversations, keeping tone and expression consistent across a full exchange with latency as low as 80ms.

Unified Voice Agent API

A single API combines speech-to-text, LLM orchestration and text-to-speech, handling conversation context, prompt updates, response injections and function calling so you don't have to glue separate services together.

Audio Intelligence

Audio Intelligence endpoints pull insights out of recorded or live conversations, though concurrency is capped at 10 REST requests on the self-serve plans.

Flexible deployment

You can run models in real time or batch, and in Deepgram's cloud or self-hosted, which gives regulated teams an option that pure SaaS rivals often don't.

Telephony and transport partners

Partner-provided transport layers stream audio into Deepgram and play TTS back to callers, including telephony connections over PSTN and SIP.

Best for

developers building real-time voice agents
startups adding transcription to a product, like Granola did for meeting notes
voice agent platforms that need high-concurrency STT and TTS
enterprises that need self-hosted speech models
teams tired of stitching together separate STT, LLM and TTS vendors

Skip if

Skip it if you want a ready-made transcription or voiceover app. Deepgram is an API platform and you'll need to write code.
Skip it if you need hands-on support without an Enterprise contract. Pay As You Go and Growth only get Community and Discord support.
Skip it if your workload is mostly Whisper-based, since Deepgram Whisper Cloud is capped at 5 concurrent requests even on Growth.

Pros & cons

Pros

The free tier gives you $200 in credit with no credit card, no minimums and no expiration.
Flux STT handles turn-taking and interruptions natively, which is the hardest part of making a voice agent feel natural.
One Voice Agent API covers STT, LLM orchestration and TTS, cutting integration work and round-trip latency.
Self-hosted deployment is offered alongside cloud, a real advantage for compliance-heavy teams.
Serious production users like Granola, Vapi, Daily and Cloudflare publicly build on it.

Cons

Support on self-serve plans is limited to Community and Discord with standard uptime SLAs. Dedicated support means talking to sales.
The Growth plan starts at $4K+ per year in prepaid credits, a big jump if you're between hobby usage and enterprise scale.
Some concurrency caps are tight: 5 for Deepgram Whisper Cloud and 10 for Audio Intelligence, identical across Pay As You Go and Growth.
The pricing overview lists tiers and concurrency, not per-minute rates by model, so you'll have to dig into model-level pricing to estimate real costs.

Pricing

Tier	Price	Includes
Pay As You Go	Free $200 credit, then pay-as-you-go	All endpoints in public models, no minimums, no expiration, no credit card required. STT concurrency up to 50 REST and 150 WSS, 5 for Deepgram Whisper Cloud. TTS and Voice Agent API up to 45. Audio Intelligence up to 10. Community and Discord support, standard uptime SLA.
Growth	$4K+ / year	Prepaid annual credits redeemed against usage, saving up to 20%. All endpoints in public models. STT concurrency up to 50 REST and 225 WSS, 5 for Deepgram Whisper Cloud. TTS and Voice Agent API up to 60. Audio Intelligence up to 10. Community and Discord support, standard uptime SLA.
Enterprise	Contact sales	For businesses with large volumes, specific data or deployment requirements, or support needs beyond the self-serve plans.

Frequently asked questions

Is Deepgram free to try?

Yes. The Pay As You Go plan starts with $200 in free credit, no credit card required, and the credit doesn't expire. After that you pay for actual usage.

What does the Growth plan add over Pay As You Go?

Growth starts at $4K+ per year in prepaid credits and saves you up to 20%. It also raises concurrency, for example from 150 to 225 WebSocket streams for speech-to-text and from 45 to 60 for text-to-speech and the Voice Agent API.

How is Deepgram different from using OpenAI Whisper for transcription?

Whisper is a general transcription model, while Deepgram's Flux STT is built to detect end-of-turn and interruptions in live conversation. Deepgram does offer Deepgram Whisper Cloud, but it's capped at 5 concurrent requests on self-serve plans.

Can I use Deepgram for phone-based voice agents?

Yes. The Voice Agent API works with partner transport layers, including telephony partners over PSTN and SIP, and platforms like Vapi build phone agents on Deepgram's APIs.

Can Deepgram be self-hosted?

Yes. Deepgram says its models run in real time or batch, either in its cloud or self-hosted. Custom models and specialized deployments go through the Enterprise sales team.

How Deepgram compares

Deepgram vs OpenAI Whisper

Whisper is a solid general transcription model, but Deepgram's Flux STT is the better pick for live voice agents because it's built to detect turn-taking and interruptions in real time.

Deepgram vs ElevenLabs

ElevenLabs is known mainly for voice generation, while Deepgram makes more sense if you want speech recognition, TTS and agent orchestration from one API.

Deepgram vs AssemblyAI

AssemblyAI is a direct speech-to-text API competitor, so test accuracy on your own audio. Deepgram stands out if you also want TTS, a unified Voice Agent API and a self-hosted option.

Reviews

No reviews yet, be the first.

More like DeepgramAll voice tools

Own this tool?Claim the listing to get a verified badge and edit the details.