We use Deepgram's low-latency realtime speech recognition model, Nova2. Their new Aura voices are also available on Vapi.

Generous free tier, blazing fast speeds, tts and stt I can count on.

Martin uses Deepgram's fast and accurate speech-to-text engine, Nova-2.

I wanted to allow 5-hour uploads without chunking. Then I found Deepgram. What I saw there was very fascinating.

We use Deepgram to transcribe the input audio from the user and turn it into text to feed into our agent. It's incredibly accurate and very easy to use!