Imagine the following scenario. When a patient calls a leading pharmacy, all calls are handled by an AI voice agent. The caller might switch between Spanish and English, interrupt the agent, or say the name of a complicated drug, and the AI will follow suit. For many companies, this is the reality today, and technology resolves more than 90% of such calls without escalating them to a representative.
Even in the online era, voice remains the primary way people connect with businesses. Addressing this without expanding call centers is what DeepGram calls the “voice AI economy,” the proliferation of automated conversation technology into commercial hubs like pharmacies, call centers, and even air traffic control.
A strong technical foundation is essential to support critical applications and, perhaps counterintuitively, goes well beyond traditional speech-to-speech and text-to-speech models. In the case of Deepgram, the foundation is provided by Nvidia’s Dell AI Factory. Deepgram builds voice technology and AI Factory powers it.
“Voice economy” indicators are performing well
Kris Efland is VP of Engineering at Deepgram. He believes the company’s core value proposition is as simple as casual conversation itself: the power of the word spoken and the power of being understood.
“The biggest value created by AI is its deep integration into everyday life,” Efland said. But the seemingly casual nature of UX Deepgram’s functionality is not without its complexities, he said. “It’s clearer than ever that orchestration matters and context matters.”
Deepgram holds the world’s first patent for deep learning AI for audio. Neural networks have been applied to raw sound for years before the current wave, but they have now gone far beyond their first use case: transcription tools.
Currently, AI voice agents handle regulated high-stakes tasks. You can take complete control of the call, or you can listen to the live call and provide suggestions via the call center agent’s handset, and can analyze the caller’s sentiment to return a response or recommend an escalation to a supervisor.
In the public safety field, DeepGram’s air traffic control model can track conversations even in ambient noise and when multiple people are talking at the same time. In the healthcare field, medical records can be run on a single device, securely integrating with medical records apps on the cloud while ensuring patient data never leaves the laptop.
Conversation accuracy at human speed
To provide a reliable and responsive experience, Deepgram’s AI model, Flux, responds in 200-300 milliseconds. This is fast enough that the AI agent can maintain the conversation even if the caller interrupts or loses their train of thought. Flux also provides a real-time transcript to the human agent who may lead the call. For example, when a product is mentioned, Flux instantly displays the corresponding spec sheet.
Unlike many modern AI tools, Deepgram does not rely on large-scale language models (LLMs). Instead, run specialized AI models for specific jobs and deploy them wherever you need them, either in the cloud or on local edge servers. For healthcare and financial customers who cannot publish their data on the internet, we have fully “air-gapped” systems where the model and data are completely self-contained.
Range, low latency, accuracy, and privacy are a combination not many companies can offer.
Extending voice agents through Dell AI Factory with Nvidia
Deepgram operates its own data centers and trains its own underlying models built by a team of over 30 researchers. Their work relies on Nvidia’s Dell AI Factory, which leverages Dell PowerEdge XE series servers with Nvidia accelerated computing and flexible, secure Dell PowerScale storage to build, deploy, and scale AI workloads.
Return on investment is important to Efland. Dell AI Factory, in conjunction with Nvidia, is an end-to-end enterprise AI solution that uses Deepgram’s existing data centers so his team can prioritize innovation. This allows you to add compute without building and operating new facilities, where costs drive profits.
“Dell AI Factory and Nvidia are the best approach to building capacity at the data center level,” Efland said, citing the per-cage benefits the company is realizing. “This gives us the ability to really scale and gives us a consistent infrastructure partner across the spectrum from lab and research to production engineering.”
And what’s in that production pipeline? Advances in energy and tempo matching, Efland said, and the “magical experience” where voice AI meets a human speaker in real life will prove that the “voice AI economy” is more than just a buzz, it’s an inflection point in human-AI interaction.
See how Dell AI Factory and Nvidia power real-time voice AI at scale.
this Sponsored Post Created by BI Studios and Dell AI Factory and Nvidia.