With live conversations requiring real-time human interaction, enterprises still struggle with unreliable speech recognition due to high latency, inaccuracies, interruptions, and scalability issues. Deepgram addresses this with its real-time API platform, delivering speech-to-text (STT), text-to-speech (TTS), and speech-to-speech (STS) capabilities via models such as Nova-3, Aura-2, Flux, Voice Agent API, and Saga.
Today, Deepgram raised $130M in Series C funding at a $1.3B valuation, led by AVP, with participation from existing investors Alkeon, BlackRock, In-Q-Tel, Madrona, Tiger, Wing, and Y Combinator, as well as Alumni Ventures, Princeville Capital, Twilio, ServiceNow, SAP, Citi Ventures, and Columbia University.
The capital will accelerate Deepgram’s platform, launch “Powered by Deepgram,” fund the OfOne acquisition, expand its patent portfolio, and open a new Voice AI Collaboration Hub in San Francisco.
End-to-end deep learning for voice
Co-founded in 2016 by Scott Stephenson, Deepgram emerged from frustration with legacy speech tech that couldn’t handle real-world conversations. The startup aims to build reliable, real-time voice technology that developers and enterprises can deploy, processing over 125,000 years of audio and 1 trillion words to date.
Deepgram’s tech stack features enterprise-grade, customisable models that can be deployed via cloud APIs, self-hosted, or on-premises, with full SDKs for rapid development. Key offerings include Nova-3 (the world’s most accurate real-time STT), Aura-2 (professional, cost-effective TTS), Flux (the first model that solves interruptions in voice agents), Voice Agent API (enterprise-ready conversational AI), and Saga (Voice OS).
Other features include sub-second latency, 95%+ containment in real-world use (e.g., via OfOne integration for restaurants), domain-specific customisation, and hardware-efficient processing, backed by patents such as US 12,380,880 (end-to-end ASR with transformers) and US 12,334,075 (efficient batching).
Unlike competitors such as AssemblyAI, Speechmatics, and Google Cloud Speech-to-Text, Deepgram excels at fully conversational, low-latency STS, leveraging proprietary deep learning to integrate STT/TTS seamlessly.
Over 1,300 organisations, including Twilio and ServiceNow partners, build on it for superior reliability.
What about diversity?
When we asked Deepgram about diversity, Stephenson commented, “One of Deepgram’s values is: Every voice, heard and understood. We thrive on a concert of unique voices. To create a more inclusive company, we actively welcome diverse backgrounds, thoughts, and opinions. We also support ERGs (employee resource groups) to foster connection and mutual support among Deepgrammers.
What’s next?
Deepgram will use the funding to scale globally (e.g., Europe via AVP), enhance real-time frontier models for billions of conversations, and roll out Deepgram for Restaurants with OfOne’s 95%+ containment tech for drive-thrus.
Stephenson concluded, “Over the next three to five years, Deepgram wants to cement itself as the strategic real-time Voice AI platform of record, grow a large ecosystem of ‘powered by Deepgram’ partners and products, and help move voice from a historically disappointing interface to the default way people interact with software and services.”
