Welcome (back) to AI Minds, a newsletter about the brainy and sometimes zany world of voice AI, brought to you by the Deepgram editorial team.

In this edition:

• 📺  Video: Building Voice Agents Just Got Easier

• 🎶 SingNet: Towards a Large-Scale, Diverse, and In-the-Wild Singing Voice Dataset

• 🔊 How AI Voice Morphing Reveals the Boundaries of Auditory Self-Recognition

• 🧠 Evaluating End-of-Turn (Turn Detection) Models: The Flux Chronicles

• 🧳 Enterprise Voice AI Use Cases That Deliver Measurable ROI

• 🐝 Social Media Buzz: Voice AI Gone Too Far? Deepfaking Elon Musk for 3 Months 

• 👂 Heard on X: AI Agents are already transforming how we work

• 🏆 Director of Stanford’s Institute for Human-Centered AI, Fei-Fei Li, wins QE Prize

• 🤖 In Case You Missed It: The Scott Stephenson AI Show (Episode 6)

Thanks for letting us crash your inbox; let’s party. 🎉

📺  Video: Building Voice Agents Just Got Easier

Anoop Dawar from Deepgram explores the evolution of voice AI and its real-world impact on business. He introduces Deepgram’s innovations in transcription, speech generation, and voice agents—highlighting Flux, the world’s first conversational speech recognition system.

The talk covers challenges like latency and orchestration, while envisioning a future of human-like, emotionally aware voice interactions that move closer to passing the audio Turing Test.

🔍 Research: How to train AI to sing like a human and how good humans are at recognizing their own AI-modified voices

SingNet: Towards a Large-Scale, Diverse, and In-the-Wild Singing Voice Dataset - This paper presents SingNet, an extensive, diverse, and in-the-wild singing voice dataset. Specifically, the authors propose a data processing pipeline to extract ready-to-use training data from sample packs and songs on the internet, forming 3000 hours of singing voices in various languages and styles.

VoiceMorph: How AI Voice Morphing Reveals the Boundaries of Auditory Self-Recognition - Can people still recognize their own voices even after it’s been modified by AI? This study investigated auditory self-recognition boundaries using AI voice morphing technology, examining when individuals cease recognizing their own voice.

⚡The challenge of evaluating end-of-turn and how enterprises are using voice AI today

Evaluating End-of-Turn (Turn Detection) Models - We at Deepgram were unsatisfied with the approaches to turn detection evaluation we’d seen out there, so we built our own. This post describes the data we used, why we think that’s the right data to use, and the algorithm we developed to calculate maximally precise accuracy metrics.

Enterprise Voice AI Use Cases That Deliver Measurable ROI - See 10 proven enterprise use cases of voice AI that deliver ROI across industries. From healthcare clinical documentation to intelligent appointment scheduling, the practical applications of this technology stretch far and wide!

🐝 Social Media Buzz: Voice AI Gone Too Far? Deepfaking Elon Musk for 3 Months (and more!)

🤖 In Case You Missed It: The Scott Stephenson AI Show (Episode 6)

“Google's $24 Billion infrastructure plan, Meta's puzzling simultaneous hiring-and-layoff strategy, and OpenAI's controversial new browser. Check out a tech CEO's thoughts on the latest AI developments in this episode of "The Scott Stephenson AI Show"!