Deepgram is a fast, accurate speech-to-text (and text-to-speech) API used for transcription, captions and voice agents. It bills by the minute of audio processed, with a generous free credit to start. Here's what it costs and how to get your key.
| Model | ≈ $/audio minute | Free credit | Best for |
|---|---|---|---|
| Nova (pre-recorded STT) cheapest | $0.0043 | $200 | Best accuracy, transcription |
| Nova (streaming STT) | $0.0059 | $200 | Live captions, voice agents |
| Aura (text-to-speech) | ~$0.015 / 1k chars | $200 | Voice output |
→ Adding voice to an app? Estimate the LLM side with the AI app cost estimator.
Deepgram gives $200 in free credit on signup — enough to transcribe hundreds of hours — with no credit card required to start. After that it's pay-as-you-go per minute. That free credit makes it one of the easier speech APIs to trial.
1. Sign up at deepgram.com (free $200 credit).
2. Open the Console → API Keys and create a key.
3. Send audio by URL or file upload to the transcription endpoint.
4. Add billing only when you exhaust the free credit.
Test it with a simple request:
AssemblyAI and OpenAI Whisper are the main rivals for speech-to-text (Whisper is ~$0.006/min); Google and Amazon Transcribe also compete. For voice output, ElevenLabs leads on realism.
Speech-to-text is around $0.0043 per audio minute on the Nova model — about $0.26 per hour. Streaming is slightly more. New accounts get $200 free credit.
Yes — $200 in free credit on signup, no card required, then pay-as-you-go.
All three are strong; Deepgram is known for speed and the big free credit, Whisper (via OpenAI) is cheap and simple, AssemblyAI adds audio-intelligence features.
Sign up at deepgram.com, open Console → API Keys, and create one.
Not affiliated with Deepgram. Prices are reference estimates — always verify on the official pricing page.