Imagine this scenario: A patient calls a leading commercial pharmacy and the call is handled entirely by an AI voice agent. The caller might switch between Spanish and English, interrupt the agent, and cite complicated drug names, but the AI follows it all. For many businesses, this is a reality today, and the technology resolves more than 90% of such calls without escalating to a person.
Even in the online age, voice remains the primary way people connect with companies. Handling that without ever-expanding call centers is what Deepgram calls the “voice AI economy” — the spread of automated conversational technology into commercial hubs such as pharmacies, call centers, and even air traffic control.
To support critical applications, a strong technological foundation is vital and, perhaps counterintuitively, goes far beyond traditional speech-to-text and text-to-speech models. For Deepgram, that foundation is provided by the Dell AI Factory with Nvidia. Deepgram builds the voice technology, and the AI Factory powers it.
‘Voice economy’ indicators are strong
Kris Efland is vice president of engineering for Deepgram. He sees the company’s leading value proposition as simple as, well, casual conversation itself: the power of the spoken word, and its power to be understood.
“The most value that’s being generated by AI is deep integration into the day-to-day,” Efland said. But the casual-seeming nature of the UX Deepgram powers is not without complexity, he said. “It’s clearer than ever that orchestration matters, and context matters.”
Deepgram holds the world’s first patent on deep learning AI for audio. It was applying neural networks to raw sound years before the current wave, and has now moved well beyond the transcription tools that were its first use case.
Today, AI voice agents handle regulated, high-stakes work. They can manage calls entirely or listen to live calls and offer suggestions via the call center agent’s earpiece, providing responses or recommending escalation to a supervisor by analyzing the caller’s emotions.
In public safety, Deepgram’s air-traffic control model can track conversations even amid background noise and when several people are talking at once. In healthcare, it can run medical transcriptions on a single device, so patient data never leaves the laptop while securely integrating with medical records apps in the cloud.
Conversational accuracy at human speed
To deliver a reliable, responsive experience, Deepgram’s AI model, Flux, replies in 200 to 300 milliseconds — fast enough that if a caller interjects or loses their chain of thought, the AI agent still holds the thread of their conversation. Flux even supplies a real-time transcript to a human agent who might be leading the call. For example, if a product is mentioned, Flux immediately surfaces the corresponding spec sheet.
Unlike many contemporary AI tools, Deepgram does not depend on large language models (LLMs). Instead, it runs specialized AI models for specific jobs and deploys them where needed, whether in the cloud or on local edge servers. For healthcare and financial customers whose data cannot be on the internet, there are fully “air-gapped” systems in which the model and data are entirely self-contained.
Range, low latency, accuracy, and privacy are a combination that many businesses cannot offer.
Scaling voice agents via the Dell AI Factory with Nvidia
Deepgram runs its own data centers and trains its own foundational models, built by a team of more than 30 researchers. Their work depends on the Dell AI Factory with Nvidia, utilizing Dell PowerEdge XE-Series servers with Nvidia-accelerated computing, along with flexible, secure Dell PowerScale storage to build, deploy, and scale AI workloads.
For Efland, the return on investment is about focus. His team can prioritize innovation because the Dell AI Factory with Nvidia is an end-to-end enterprise AI solution that uses Deepgram’s existing data centers, so they can add compute without building and running new facilities, where costs would eat into margins.
“The Dell AI Factory with Nvidia is a best-of-breed approach that lets us build capacity at the data-center level,” Efland said, citing the per-cage benefit the company realizes. “It gives us the ability to really scale, and a consistent infrastructure partner across the full gamut for the lab, research, all the way to production engineering.”
And what’s in that production pipeline? Advances in energy and tempo matching, where voice AI truly meets human speakers where they are — “magical experiences,” Efland said, that prove the “voice AI economy” isn’t just a talking point, it’s an inflection point in human-to-AI interaction.
Explore how the Dell AI Factory with Nvidia powers real-time voice AI at scale.
This sponsored post was created by BI Studios with the Dell AI Factory with Nvidia.
