Deepgram logo

Deepgram

Developer-first voice AI platform for speech-to-text, text-to-speech, and voice agents at scale.

★ 4.7 / 5FreemiumVoice
Visit Deepgram →
Free planYes, $200 in free credit, no card required, and it does not expire.
Paid plans fromPay as you go after the free credit; Growth from $4,000 a year in pre-paid credits.
Runs onCloud API, with self-hosted containers on Enterprise.
Websitedeepgram.com

Deepgram is a voice AI platform built for developers and businesses that need speech turned into text, text turned into speech, or a full voice agent that can hold a phone conversation. It is an API first product: you send audio or text to Deepgram and get results back in real time, rather than working inside a recording studio style app.

What sets Deepgram apart is speed and price at volume. Its Nova-3 transcription model supports more than 60 languages and bills per second of audio with no rounding up, and its Flux models are built for live voice agents where every millisecond of delay matters. More than 650,000 developers and 5,000 organisations build on it, and the platform reports over 200,000 years of audio processed.

If you are building a product that listens or talks, Deepgram is one of the most cost effective places to start, and the $200 free credit means you can prototype without a card.

Features

Nova-3 speech-to-text

Deepgram's flagship transcription model, covering more than 60 languages with speaker diarization, smart formatting, and keyterm prompting. Streaming transcription costs $0.0048 a minute on pay as you go, or about $0.46 an hour.

Flux models for live voice agents

Flux speech-to-text and text-to-speech are built for real-time conversation, with low latency and streaming output so a voice agent can answer while the caller is still speaking. Flux STT English costs $0.0065 a minute on pay as you go.

Text-to-speech voices

Flux TTS and the Aura voice family turn text into natural sounding speech, billed per 1,000 characters of input text. Rates differ by model and plan, and the pricing page lists the current figures.

Voice Agent API

One managed API that combines speech-to-text, a language model, and text-to-speech into a working voice agent, billed per minute of WebSocket connection time. Bringing your own language model or voice moves you to a lower rate.

Audio Intelligence add-ons

Optional extras on top of transcription, including redaction ($0.0020 a minute), speaker diarization ($0.0020 a minute), entity detection ($0.0017 a minute), and keyterm prompting ($0.0013 a minute). Smart formatting is included at no cost.

Enterprise security and self-hosting

SOC 2 Type 1 and 2 certified, HIPAA compliant with a business associate agreement on Enterprise, GDPR ready with a dedicated EU endpoint, and self-hosted containers for teams that cannot send audio to the cloud.

Use cases

  • Developers add transcription, captioning, or voice input to an app with a few API calls.
  • Contact centres transcribe calls in real time and analyse them for quality and compliance.
  • Product teams build voice agents that answer calls, book appointments, or triage support.
  • Media and podcast teams transcribe long recordings and make them searchable.
  • Healthcare and finance teams use the Enterprise tier for compliant, private voice processing.

Pros and cons

What’s good

  • Very low per-minute transcription prices, with billing per second and no rounding up.
  • $200 in free credit with no card required, enough for roughly 700 hours of Nova-3 transcription.
  • Fast streaming models built for live voice agents, not just batch files.
  • More than 60 languages, plus diarization, redaction, and smart formatting.
  • Strong compliance story: SOC 2, HIPAA, GDPR, EU endpoint, and self-hosting on Enterprise.
  • No minimums on pay as you go, and volume discounts on Growth and Enterprise.

What isn’t

  • It is an API for developers, not a simple app for non-technical users.
  • Text-to-speech rates are not shown as plainly as the transcription rates on the pricing page.
  • Add-ons like diarization and redaction are billed on top of transcription.
  • Going over your Growth credits bills at your rate plus 10 percent unless you renew early.

Pricing

  • Pay as you go, $0 to start with $200 free credit: no minimums, no card required, and the credit does not expire. Speech-to-text from $0.0048 a minute (Nova-3 monolingual) or $0.0065 a minute (Flux STT English); multilingual Nova-3 is $0.0058 a minute. Text-to-speech is billed per 1,000 characters, with rates by model on the pricing page.
  • Growth, from $4,000 a year: pre-paid annual credits redeemed against usage, saving up to 20 percent. Nova-3 monolingual drops to $0.0042 a minute and Flux STT English to $0.0057 a minute, with higher streaming concurrency (225 connections versus 150 on pay as you go).
  • Enterprise, custom pricing: for large volumes, custom models, dedicated support, self-hosted deployment, and a named account team. Volume pricing ladders down with committed usage.

The $200 free credit covers roughly 43,000 minutes of Nova-3 transcription. Deepgram bills per second of processed audio with no minimum duration, and multichannel audio bills by total duration across channels.

Deepgram changes its model lineup and promo rates regularly, verify current details at deepgram.com before subscribing.

How it works

  1. Create a free Deepgram account (no card required) and claim the $200 credit.
  2. Create an API key in the Console and pick a model, such as Nova-3 for transcription or Flux for live agents.
  3. Send audio to the speech-to-text API or text to the text-to-speech API.
  4. Add extras only if you need them, such as diarization, redaction, or keyterm prompting.
  5. Move to Growth when your volume passes $4,000 a year in usage, or talk to sales for Enterprise.

Who it’s for

  • Developers building transcription, captioning, or voice features into a product.
  • Teams building live voice agents where latency and cost per minute matter.
  • Contact centres that need accurate real-time call transcription at scale.
  • Media and podcast teams transcribing long recordings on a budget.
  • Regulated teams that need HIPAA, GDPR, or self-hosted voice processing.

Final verdict

Deepgram is one of the best value voice AI APIs available: fast streaming transcription, honest per-second billing, and a $200 free credit that lets you build a real prototype before spending anything. Its weaknesses are that it is developer focused, so non-technical users will find it harder than a studio style tool, and its text-to-speech pricing is less transparent than its transcription pricing.

Choose Deepgram if you are building a product that listens or talks and you want low cost per minute, real-time speed, and strong compliance options.

Choose something else if you want a simple no-code voice cloning or narration app (ElevenLabs or Murf), or if you need a finished contact centre product rather than an API to build one.

Related reading: Best AI Voice Generators for Beginners (Free vs Paid)

Questions people ask

Is Deepgram free to use?
You can start free: every new account gets $200 in credit with no card required, and it does not expire. That covers roughly 43,000 minutes of Nova-3 transcription. After the credit runs out you pay per use, with no monthly subscription on pay as you go.
How much does Deepgram speech-to-text cost?
On pay as you go, Nova-3 monolingual streaming is $0.0048 a minute (about $0.46 an hour) and Flux STT English is $0.0065 a minute. Multilingual models cost a little more. Growth plans from $4,000 a year cut those rates by up to 20 percent.
How is Deepgram text-to-speech billed?
Per 1,000 characters of input text, across Flux TTS, Aura-2, and Aura-1. Rates differ by model and plan, and the pricing page lists the current figures.
Does Deepgram offer a voice agent API?
Yes. The Voice Agent API combines speech-to-text, a managed language model, and text-to-speech in one connection, billed per minute of WebSocket connection time. Bringing your own language model or voice lowers the rate.
Is Deepgram HIPAA compliant?
Yes. Deepgram is SOC 2 Type 1 and 2 certified, HIPAA compliant with a business associate agreement for Enterprise customers, GDPR ready with a dedicated EU endpoint, and CCPA and PCI compliant.
Can I self-host Deepgram?
Yes, on Enterprise. Self-hosted containers run in your own cloud or on-premises hardware and need NVIDIA GPUs. It is aimed at teams with data sovereignty or security requirements that rule out cloud processing.

More Voice tools

ElevenLabs logo

ElevenLabs

Voice
★ 4.9

AI voice platform — realistic text-to-speech, voice cloning, dubbing, and voice agents.

Speechify logo

Speechify

Voice
★ 4.7

Text-to-speech app that reads anything aloud in natural AI voices, up to 5x speed.

Cartesia AI logo

Cartesia AI

Voice
★ 4.6

Real-time voice AI platform with Sonic text-to-speech, Ink transcription, and phone voice agents.

Murf AI logo

Murf AI

Voice
★ 4.6

Murf AI review for 2026 — 150+ voices in 35+ languages, Studio, Dub, voice changer, Creator from $19 a month, and the API.

Prices change — always check the vendor’s site before buying. Found a price that’s out of date? Tell us and we’ll fix it.