← All organizations

Voice AI Infrastructure

Deepgram

Deepgram builds speech recognition, text-to-speech and voice agent APIs for developers and enterprises. Teams use them to transcribe live streams and recordings or build applications that listen and respond. Deepgram grew out of an audio search problem. After working on dark matter at the University of Michigan, co-founders Scott Stephenson and Noah Shutty built a tool to find specific moments in hundreds of hours of personal audio recordings. Stephenson recounted the project in Deepgram’s 2018 launch post.

In a live conversation, a pause can mean someone is thinking or that they’ve finished speaking. Deepgram’s Flux speech recognition model transcribes streaming audio and signals when it detects the end of a turn. Developers can require more confidence before Flux sends that signal, trading a slightly longer wait for fewer premature replies. The interactive avatar company Anam runs Flux on its own Kubernetes cluster and uses those signals to time its avatars’ responses.

Deepgram has expanded into both ready-made voice applications and infrastructure for teams building their own. In January 2026, it acquired OfOne, whose drive-through voice technology became the basis for Deepgram for Restaurants. In October 2026, Deepgram announced it was acquiring Daily, the company behind Pipecat. The planned platform would combine Deepgram’s speech models with Daily’s network for carrying live audio and Pipecat’s framework for connecting models and services. Deepgram said Pipecat would remain open source and vendor-neutral and that the companies were establishing an independent Pipecat Foundation.

2 talks

Newest first

Loading more talks and speakers…

2 speakers at AIE

Affiliations reflect their AIE event listings and recorded appearances, not necessarily current employment.

Company sources · checked 2026-11-13