Picture a situation where a patient reaches out to a prominent pharmacy, and the entire conversation is managed by an AI voice assistant. The interaction might involve switching between Spanish and English, interruptions, or referencing intricate medication names, yet the AI handles it seamlessly. This scenario is a present-day reality for numerous businesses, with advanced technology resolving over 90% of such inquiries without needing human intervention.
Even in a digitally driven world, voice communication remains the primary method through which individuals connect with organizations. Managing this without continually expanding call centers is what Deepgram refers to as the "voice AI economy," which involves the implementation of automated conversational technologies in commercial sectors like pharmacies, customer service centers, and even air traffic control systems.
A robust technological framework is crucial for supporting these essential applications. Interestingly, it extends far beyond conventional speech recognition and generation models. For Deepgram, this foundation is supplied by the Dell AI Factory in collaboration with Nvidia, where Deepgram develops the voice technology, bolstered by the capabilities of the AI Factory.
Strong Indicators for the 'Voice Economy'
Kris Efland, Deepgram’s Vice President of Engineering, believes that the company's primary value proposition is as straightforward as informal conversation — the marvel of human speech and the ability to comprehend it effectively.
"The greatest value from AI emerges from its deep integration into everyday tasks," Efland explained. However, he notes that the seemingly simplistic user experience facilitated by Deepgram is inherently complex, emphasizing the importance of orchestration and context.
Deepgram possesses the world’s first patent for deep learning AI related to audio processing, having explored neural networks on raw sound long before the current trend. It has now evolved far beyond the transcription tools that marked its initial application.
Today, AI voice agents engage in critical, regulated tasks. They can independently manage calls or listen in on live conversations, providing real-time suggestions to call center representatives through discreet communications, analyzing callers' emotions to recommend appropriate responses or escalate issues to a supervisor.
In the realm of public safety, Deepgram's air traffic control model can discern conversations even in noisy environments or when multiple people speak simultaneously. In healthcare, it facilitates medical transcription on a single device, ensuring patient data remains contained while securely connecting with cloud-based medical record applications.
Human-like Conversational Precision
To ensure a fast and effective experience, Deepgram’s Flux AI model responds within 200 to 300 milliseconds — quick enough to keep up with a caller's interjections or shifts in thought. Additionally, Flux provides a real-time transcript to assist any human agent supervising the call. If a product is referenced, Flux can instantly retrieve the relevant specification sheet.
Unlike many existing AI solutions, Deepgram does not rely on large language models (LLMs). It employs specialized AI models tailored for specific tasks and deploys them as needed, whether in the cloud or on local edge servers. For sectors like healthcare and finance, where data security is paramount, Deepgram offers fully "air-gapped" systems that maintain complete data and model isolation.
The combination of range, low latency, accuracy, and privacy is something many businesses struggle to provide.
Scaling AI Voice Agents through the Dell AI Factory with Nvidia
Deepgram operates its own data centers and develops its fundamental models with the help of over 30 researchers. Their progress heavily relies on the Dell AI Factory in tandem with Nvidia, utilizing Dell PowerEdge XE-Series servers equipped with Nvidia’s accelerated computing, along with adaptable and secure Dell PowerScale storage to develop, launch, and scale AI tasks.
According to Efland, the focus shifts to innovation through this partnership. The Dell AI Factory with Nvidia serves as a comprehensive AI solution that integrates with Deepgram’s existing data centers, enabling them to enhance computational capacity without the need to construct and maintain additional facilities, where expenses could impact profit margins.
"The Dell AI Factory with Nvidia represents a leading strategy that allows us to expand capacity at the data center level," Efland noted, highlighting the benefits realized in their operations. "It empowers us to scale efficiently and provides a consistent infrastructure partner across all phases, from research and development to production engineering."
So, what’s on the horizon for production? Advancements in energy and tempo matching, where voice AI genuinely aligns with human speech patterns, creating "magical experiences," according to Efland. It’s clear that the "voice AI economy" is not merely a buzzword, but a significant turning point in human-AI interactions.
Discover how the Dell AI Factory with Nvidia is enabling real-time voice AI at scale.
