Deepgram

Overview

Deepgram provides voice AI infrastructure for developers: speech-to-text, text-to-speech and a Voice Agent API for building assistants that listen and reply in real time. It is aimed at products such as call centres, voice agents, meeting tools and media transcription.

Main features

• Speech-to-text: Nova-3 models for streaming and pre-recorded audio in monolingual and multilingual versions, and Flux, a conversational model built for real-time dialogue. Whisper Large is also available.
• Text-to-speech: Aura-2, Aura-1 and Flux TTS voices; Deepgram lists more than 45 languages and over 100 voice profiles.
• Voice Agent API: combines listening, a language model and speech into one agent, with options to bring your own LLM or TTS and to call functions, for example to book appointments.
• Enterprise: custom data and deployment requirements on enterprise plans.

Pricing

New accounts get $200 of free credit, with no minimum, no expiry and no credit card required. Pay-as-you-go examples:

• Speech-to-text: Nova-3 Monolingual $0.0077 per minute streaming or $0.0043 per minute pre-recorded; Flux STT English $0.0077 per minute.
• Text-to-speech: Aura-2 $0.030 per 1,000 characters; Aura-1 $0.0150.
• Voice Agent API: $0.075 per minute for the standard configuration. The Growth plan (from $4,000 a year in prepaid credit) lowers these rates by up to 20%; Enterprise is custom.

Who it is for

Developers and product teams building voice features: transcription, call analytics, voice bots and real-time assistants.

How to start

Sign up to get the free credit, create an API key, and send a short recording to the pre-recorded endpoint to compare accuracy on your own audio. Then test streaming with a live microphone before choosing a model.

Things to note

Prices differ by model, by streaming versus pre-recorded audio and by plan, so estimate with your real minutes. Test accents, names and domain terms on your own data.

Similar tools

ElevenLabs, AssemblyAI, OpenAI Whisper, Speechmatics, Rev.