We don't just build features; we invent the foundational architectures that power the future of human-machine communication. Deepgram is a "voice-native" company. Our portfolio of granted patents secures our radical, purely data-driven approach to speech-to-text, text-to-speech, and conversational agents.
Modern voice agents require more than just speed; they need context. These patents cover our unique methods for unified end-to-end model training.
Scaling to trillions of words requires extreme optimization. We patented the way we use GPUs to drive down latency and cost.
Voice data is only valuable if it is searchable and adaptable. These patents define how we index neural network "internal states" to find insights and recognize specialized jargon.
The patents below represent Deepgram’s broader portfolio, including continuations and extensions of the patent families highlighted above.
US 11,367,433
End-to-end neural networks for speech recognition and classification.
US 10,847,138
Deep learning internal state index-based search and classification.
US 10,720,151
End-to-end neural networks for speech recognition and classification.
US 10,380,997
Deep learning internal state index-based search and classification.
US 10,210,860
Augmented generalized deep learning with special vocabulary.
Join 200,000+ developers building with our patented Voice AI.
Notice of Virtual Patent Marking This page serves as notice under 35 U.S.C. § 287(a) that the Deepgram products and services listed here are protected by the associated patents and pending applications in the U.S. and other jurisdictions.