July 24 2026

Deepgram announced a new initiative to bring enterprise-grade speech recognition directly onto PCs powered by Snapdragon processors. By optimizing its Nova-3 speech-to-text model on the Qualcomm Hexagon NPU in the Snapdragon X Series platform, Deepgram is enabling developers and device manufacturers to deliver real-time voice experiences with greater speed, privacy, and reliability, without relying on a cloud connection. This effort also opens the door to further integration of voice Ai into more product categories, working purely on-device.

Most voice AI solutions that support natural language interactions have historically relied on a cloud-based architecture. Before a response could be delivered, each interaction required audio to leave the device, travel to the cloud, be processed somewhere else, and then return. With this approach, delays and privacy concerns are sometimes introduced, which limit where voice AI can realistically be deployed.

Deepgram is focusing on changing that model, enabling speech recognition to happen directly on the device itself. The result, the company says, paves the way to user experiences that feel like a natural conversation, whether it is running in a vehicle, consumer electronics, or on the latest generation of ARM-based PCs. As Deepgram explains, its Nova-3 speech-to-text model running on PCs powered by Snapdragon is able to deliver real-time voice experiences without Internet access. The company calls it enterprise-grade speech recognition without the cloud.

“We recognize that voice AI technologies serve different needs. That’s why we value our work with Deepgram – because accuracy, latency, reliability, and scalability matter for developers building automotive systems, healthcare applications, industrial solutions, and other mission-critical experiences,” says Upendra Kulkarni, Vice President, Product Management, Qualcomm Technologies. “The on-device AI capabilities of Snapdragon processors combined with Deepgram's enterprise-class voice AI uniquely bring those requirements together in a way that opens new possibilities for real-time voice experiences across the device ecosystem."

"The future of voice-first interaction is about meeting users wherever they are — in a vehicle, on a factory floor, inside an XR headset — without compromising on accuracy or responsiveness," adds Abe Pursell, Vice President, Business Development and Partnerships, Deepgram. "To make that possible, you need enterprise-grade speech recognition that can deliver within the constraints of a device. We're bringing Deepgram Nova-3 directly to the ecosystem of PCs with Snapdragon, providing a foundation to build voice into products that work wherever people happen to be, regardless of network conditions."

As Deepgram explains, the Nova-3 STT model made enough progress on accuracy, even in challenging audio conditions, to deliver this level of user interaction, running purely on-device. The company also claims that Nova-3 is the first voice AI model to offer real-time multilingual transcription; a claim other companies will dispute with significant progress on this front happening nearly every month.

Deepgram also claims to be the first to provide users with “demonstrably effective and highly accurate self-serve customization – enabling instant vocabulary adaptation without model retraining.” And reinforcing its claim of superior accuracy, the company states: “Nova-3 leads transcription accuracy with a 6.89% word error rate on real-world production audio, a 24.7% lower error rate than the next-best competitor.”

Independently of the virtues of its voice AI engine, this partnership with Qualcomm places Deepgram at the forefront of an important market trend. On-device voice interfaces are seeing strong demand in multiple product categories, starting with automotive, and Qualcomm Snapdragon PCs offer a perfect platform to prove to the market that their model delivers the goods.

A leading name in the Voice AI economy, having processed over 50,000 years of audio and transcribed over 1 trillion words, Deepgram technologies have been embraced by more than 200,000 developers and 1,400 organizations. The company’s voice AI platform offers speech-to-text (STT), text-to-speech (TTS), and full speech-to-speech (STS) capabilities, all powered by its enterprise-grade runtime. Deepgram’s voice-native foundation models, accessed through cloud APIs or as self-hosted/on-premises APIs, deliver leading accuracy, low latency, and competitive pricing.
www.deepgram.com

About Joao Martins

Since 2013, Joao Martins leads audioXpress as editor-in-chief of the US-based magazine and website, the leading audio electronics, audio product development and design publication, working also as international editor for Voice Coil