- The AI Underground With CAiS
- Posts
- The AI Voice Shift: How AI Is Learning to Talk, Listen, and Feel
The AI Voice Shift: How AI Is Learning to Talk, Listen, and Feel
Do you know how we got from a talking shoebox (i.e., chatbots) to a voice (i.e., AI voice within the agenticOS) that can narrate your memoir in 12 languages? What happens when a synthetic voice sounds more polished than your top salesperson (you might be surprised how an AI-first company like CAIS feels about this)? Is it still your voice if it speaks in a language you've never learned or saying things you had not originally conversationally thought of on the fly? When does automation become intrusion, and can a robocall ever be respectful? When does AI voice become illegal and associated with costly fines? Will the next voice you trust most... be artificial? Explore this and much more in the newsletter below!
Ross Green
May 20, 2025
In partnership with
Table of Contents
Introduction:
What if your doctor could offer calm reassurance without being in the room? What if your podcast could speak in five languages, and still sound like you? And what if your customer service team never slept, never misspoke, and always remembered your brand’s tone?
These aren’t distant dreams; thanks to rapid advances in synthetic voice technology, they’re already emerging as everyday realities. Voice is becoming an entire platform for how we relate to machines, content, and each other.
Enter the agenticOS with its definition here.
We’ve entered a transformative moment for voice AI. No longer confined to stilted GPS voices or awkward call center bots, today’s systems can listen like attentive partners and speak with startling realism, even emotion to a certain extent. Check our Custom AI Studio (CAIS)’s very own AI agent demo here.
In this newsletter, we trace how we got where we are today with AI voice technology, where we are now, and where this tech is going next. Along the way, we’ll explore the breakthroughs, use cases, objective datat, controversies, and emerging frontiers that are shaping the sound of the synthetic voice future.
We begin in Section 1: Foundations of AI Voice, which explores how this field evolved. In Section 1.1: Early Innovations, we examine the mechanical beginnings of speech synthesis and recognition, from Bell Labs’ Voder and Audrey to IBM’s Shoebox and DARPA’s early research. Then, in Section 1.2: The Deep Learning Revolution, we follow the shift from rule-based systems to neural networks, showcasing models like Whisper, WaveNet, and Tacotron that laid the groundwork for today’s incredibly lifelike voices.
In Section 2 we apply AI voice into the agenticOS and how it works today. As such, Section 2: Real-World Use Cases of AI Voice looks at where these breakthroughs are making a difference in the world today. In Section 2.1: AI Voice for Inbound Support & Sales, you’ll see how smart IVRs and brand-customized agents are reshaping customer service, with use cases from Wendy’s drive-thrus to multilingual virtual receptionists. Section 2.2: AI Voice in Content Creation unpacks how audiobooks, podcasts, and global video dubbing are being transformed by synthetic voice tech, thereby lowering costs while expanding creative reach. And in Section 2.3: RVMs, Outbound Calls & Ethics, we dive into the most controversial territory: the rise of ringless voicemails, the fine line between helpful automation and spam, and the legal limits brands must now navigate.
In Section 3: AI Voice—A Table of Facts, we break from narrative and speculation to give you the hard numbers: What does voice AI cost? Who’s leading the field? How big is the market? What’s the most adopted, and most controversial, use of voice today?
Finally, we close with a forward-looking spotlight in Section 4: Quickie – Emotionally Tuned AI Voices. Here we speculate on what’s next: emotionally aware voice agents that can adapt their tone based on your mood, speak like therapists, and deliver interactive stories or lessons that change as humans respond.
Please read on and let’s explore the evolution, adoption, and emotional depth of the AI voices reshaping how we communicate.
Section 1: Foundations of AI Voice in the agenticOS
For over a century, humans have dreamed of machines that could talk, and listen, like we do. From clunky lab prototypes like the Voder and Shoebox to the smooth voices of Siri and Alexa to post-Siri AI voice (i.e., “modern AI voice”), the journey of artificial voice has been one of relentless innovation. Thanks to deep learning, today's systems can finally mimic speech with uncanny realism after decades of hard work, thereby bringing us closer than ever to natural human-computer conversation. Click here to learn more about natural language processing (NLP).
Here's how we got here:
Section 1.1: Early Innovations: Talking Machines in Labs Human beings have long been fascinated by the idea of talking machines, or devices that could mimic the sounds of speech or even understand and respond to us in kind. This ambition has driven innovation across decades, from the mechanical marvels of the early 20th century to the embedded AI assistants that live in our phones and homes today. This section explores the evolution of AI voice through two key phases: the early innovations of speech labs and the scientific breakthroughs that transformed them into intelligent voice assistants...read more here.  “The Voder — Bell Telephone Laboratories.” The Bell System Technical Journal, vol. 19, no. 4, Oct. 1940, pp. 510–514. American Telephone and Telegraph Company. Available at: https://archive.org/details/bellsystemtechni19amerrich/page/507/mode/1up. Accessed May 14, 2025.  “IBM 7094.” History of Information, www.historyofinformation.com/image.php?id=1011. Accessed May 14, 2025, by Ross W. Green, MD. | Section 1.2: Today’s Lifelike AI Voices By the 2010s, decades of rule-based and statistical voice systems began giving way to a new kind of intelligence, one fueled by data, deep learning, and raw computing power. Voice AI made its most radical leap yet, shifting from pre-programmed templates to neural models capable of learning patterns, accents, and emotional cues directly from human speech. This section explores two major breakthroughs: the rise of end-to-end deep learning in speech recognition, and the emergence of neural text-to-speech systems that synthesize truly humanlike voices…read more here.  “VALL-E X: Cross-Lingual Speech Synthesis with High-Quality Personalized Voice.” Microsoft Research, Microsoft, https://www.microsoft.com/en-us/research/project/vall-e-x/. Accessed May 14, 2025.  Ross W. Green (May 14, 2025). “Key Milestones in Deep Learning AI Voice Tech.” Napkin.ai |
Section 2: Real-World Use Cases of AI Voice with the agenticOS
AI voice is reshaping how businesses connect with customers. From smart support agents replacing phone menus to synthetic narrators powering podcasts to outbound voice campaigns raising ethical questions (and potentially hefty fines from the government), the use cases are expanding fast. In this section, we explore where AI voice is making a real impact today, on the frontlines of service, sales, and storytelling, and the considerations that come with it.
Section 2.1: AI Voice for Inbound Support & Sales w/ the agenticOS AI-driven voice assistants are rapidly transforming how companies handle incoming customer inquiries. Modern smart IVR systems leverage conversational AI to understand natural speech and intent, moving beyond rigid “press 1, press 2” menus we have all become familiar with…read more here.  Ross W. Green (May 14, 2025). “AI Voice in the U.S. Restaurant Industry” Canva.com | Section 2.2: AI Voice in Content Creation Synthetic voice technology has matured to the point that it’s becoming a practical tool for content creators, big and small. One major area of adoption is audiobook and podcast narration. Traditionally, recording an audiobook or narration meant hiring voice talent and booking studio time, representing a costly process …read more here.  “Spotify Is Launching a Voice Translation Feature for Podcasts Using AI.” Spotify Newsroom, 25 Sept. 2023, https://newsroom.spotify.com/2023-09-25/ai-voice-translation-pilot-lex-fridman-dax-shepard-steven-bartlett. Accessed May 14, 2025. | Section 2.3: RVMs, Outbound Calls & Ethics w/ the agenticOS Not all voice applications are customer-initiated given that businesses are also using AI voice for outreach. One trending technique is Ringless Voicemails (RVMs) for marketing and sales. From the client’s (sender) perspective, RVMs offer an enticing proposition…but at the risk of spamming customers associated with large fines…read more here.  Ross W. Green (May 14, 2025). “Ringless Voicemails Visually.” Canva.com |
Section 3: By the Numbers: AI Voice Technology Fact Table (2025 Update)
As of 2025, AI voice technology has reached impressive maturity, with OpenAI’s Voice Engine leading the field in text-to-speech (TTS) by enabling near-perfect voice cloning from just 15 seconds of audio. Costs remain low, typically $0.01–$0.10 per generated minute, which has fueled mass adoption—over 80% of customer service organizations now deploy voice AI, primarily for automated inbound support. Political robocalls and deepfake impersonations remain the most controversial applications, prompting new FCC restrictions. Commercially, the space is thriving: ElevenLabs, now valued at $3.3 billion, powers major publishing and gaming projects, while healthcare has emerged as the fastest-growing vertical for voice AI integration. With the global market projected to reach $8.7 billion by 2026 and startups flooding into the space, voice AI is no longer a futuristic concept—it's a foundational layer of how humans and machines now interact.
Category | Fact | Source Link |
Most advanced TTS model | OpenAI’s Voice Engine (previewed 2024) can clone a speaker’s voice from just a 15-second sample of audio, producing natural-sounding speech nearly indistinguishable from the original (currently limited preview) | |
Cost per AI-generated min | Generating AI voices costs only a few cents per minute (often roughly $0.01–$0.10 per minute, depending on provider and quality), thereby dramatically cheaper than human voice | |
Adoption rate (Customer Service) | Voice AI has quickly proliferated in customer support: over 80% of organizations now use some form of voice agent in their contact centers (from basic IVR to AI bots), though only ~21% are “very satisfied” with current solutions – spurring demand for more human-like AI | |
Top use case (2025) | Automated inbound customer service calls remain the top use case for voice AI. In leading companies, AI voice agents now handle the majority of tier-1 support calls, often boosting customer satisfaction by eliminating hold | |
Most controversial use | Political robocalls and deceptive deepfakes are widely seen as the most controversial uses of AI voice tech. For example, after an AI-cloned voice impersonated President Biden in election robocalls, regulators (FCC) moved to ban the use of AI-generated voices in scam. More on deepfakes? Check this out. | |
Market growth | The AI voice tech market is experiencing rapid growth. It reached an estimated $5.4 billion in 2024 (a 25% increase over the prior year) and is projected to grow to about $8.7 billion by 2026 | |
Leading provider | ElevenLabs – a generative voice startup – has emerged as a leading commercial provider. As of early 2025 it was valued around $3.3 billion, and it now partners with major publishers (e.g. The New Yorker, Washington Post, The Atlantic) and game studios to power AI-generated | |
Fastest-growing vertical | Healthcare is one of the fastest-growing sectors for voice AI. The voice AI healthcare market is projected to grow ~37% annually through 2030, and ~70% of healthcare organizations report that voice AI has already improved patient care in clinical. | |
Startup momentum | Voice AI is seeing a startup boom. In fact, 22% of companies in a recent Y Combinator cohort were focused on voice-based applications, thereby indicating surging investor and entrepreneur interest in voice tech. Follow the money to see into the future, friends! |
Section 4: Emotionally Tuned AI Voices for the Future of AI Voice
Ross W. Green (May 14, 2025). “Emotionally Tuned AI Voices.” Canva.com
Next-generation voice AIs are learning to speak with genuine feeling, thereby mimicking subtle emotional nuances, pacing, and even responding to listener cues rather than just the tone of the customer. Early 2025 prototypes hint at this future…read more here.
Final Thoughts on AI Voice in the agenticOS:
The story of AI voice within the agenticOS is really a story of intimacy, scale, and trust rather than a tale of technical achievement. As we’ve seen, synthetic speech has come a long way from mechanical curiosities and halting phone trees to near-human voices that can guide customers, narrate novels, and translate you into a dozen languages on, say, your podcast to reach a larger audience. These systems are already streamlining support, amplifying content creators, and prompting deep ethical questions about consent, privacy, and emotional manipulation.
What’s striking is how quickly AI voice is moving from a tool to a presence. In short, AI voice embedded within the agenticOS performs, persuades, and connects. Whether being used in a customer interaction or a bedtime story, it has the power to shape how people feel, remember, and respond. And as emotionally tuned voice agents emerge, that influence will grow even more personal and more profound.
The challenge ahead is to wield this power with intention. To ask not just what AI voice can say, but how it should say it…and why! If done right, synthetic voice will sound right, and in that future, the most human thing about our machines may be their voices.
Receive Honest News Today
Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.
Other resources:
2) Join our Community to access support from peers, a message board, and some great VIP content like our agentAcademy, weekly office hours, etc.
3) Join our weekly webinar series, "The Agentic Future with Devin Kearns" every Wednesday from 1-2 PM CST. Subscribe to this calendar for reminders.
4) Follow us on LinkedIn: Ross Green, CAiS, Devin Kearns
5) Want to learn more about how we work (e.g., build-with-you vs. build-for-you; Prebuilt SuperAgents vs, Customized Agents; etc)? Click here to schedule a meeting with us.
6) Have a friend who wants to sign-up for our Newsletter? Click here.
7) Check out Ross Green's Medium Channel
Reply