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:

• 🤯 Scott Stephenson AI Show: Why Grok is in Trouble

• 🌍 Meet Deepgram’s CTO Adam Sypniewski at two upcoming events!

• 🪡 SpeechWeave: Diverse Multilingual Synthetic Text & Audio Data Generation Pipeline

• 📈 VANPY: Voice Analysis Framework

• 🎥 Richard Sutton’s hot take: The fundamental problem with LLMs

• 🎙️ Sam Altman “speaks” Hindi with Voice AI

• 🐝 Social Media Buzz: 60ms Voice Transformation on CPU and more!

• 🔊 Google launches “Search Live”: real-time AI voice + camera search

• 👂 Study: AI-generated voices now “indistinguishable” from real 

• 📊 ElevenLabs valuation & internal share sale

• 🎙️ Scott Stephenson AI Show Podcast on Elon Musk, Geoffrey Hinton, and more!

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

🤯 Scott Stephenson AI Show: Why Grok is in Trouble and How Geoffrey Hinton sees the Future

Elon Musk's Grok is under scrutiny for generating offensive content. Meanwhile, "Godfather of AI," Geoffrey Hinton, is saying that AI will bring mass unemployment soon. And Microsoft AI CEO warns against giving AI robots rights.

🌐 Come meet Deepgram at upcoming events!

Our very own CTO Adam Sypniewski will be speaking at this event! Tech leaders (CEOs, CIOs and CTOs) register below - only 75 seats available!

When: October 16th, 3:30-6pm EST

Where: 2350 Green Rd, Suite 180, Ann Arbor, MI

Sign up here!

Our CTO Adam Sypniewski will also be speaking at the Intelligent Applications Summit 2025 The exclusive gathering for AI founders, researchers, and leaders building at the forefront of the industry and driving the AI strategies, agendas, and products of tomorrow. Sign up below!

When: September 30 - October 1, 2025

Where: Four Seasons, Seattle

Sign up here!

🔍 Multilingual Synthetic Text/Audio Data Generation and a New Voice Analysis Framework

SpeechWeave: Diverse Multilingual Synthetic Text & Audio Data Generation Pipeline for Training Text to Speech Models - High-quality Text-to-Speech (TTS) model training requires extensive and diverse text and speech data. To address these challenges, this paper proposes SpeechWeave, a synthetic speech data generation pipeline that is capable of automating the generation of multilingual, domain-specific datasets for training TTS models.

VANPY: Voice Analysis Framework - Voice data is increasingly being used in modern digital communications, yet there is still a lack of comprehensive tools for automated voice analysis and characterization. To this end, the authors of this paper developed the VANPY (Voice Analysis in Python) framework for automated pre-processing, feature extraction, and classification of voice data.

🎥 Richard Sutton’s Hot Take: Is there a “Fundamental Problem” With LLMs and Voice AI Agents?

Description: “Richard Sutton is the father of reinforcement learning, winner of the 2024 Turing Award, and author of The Bitter Lesson. And he thinks LLMs are a dead end.”

In this video, Richard Sutton is the father of reinforcement learning, winner of the 2024 Turing Award, and author of The Bitter Lesson. And he thinks LLMs are a dead end because they mimic humans rather than legitimately learning how to model the world from experience.

What do you think?

🐝 Social Media Buzz: Sam Altman “speaks” Hindi, 60ms Voice Transformation on CPU, and more!

⚡Why AI Fluency Matters — Even If You’re Not Technical and Why Banking’s Next Front Door is Voice AI

Why AI Fluency Matters — Even If You’re Not Technical - Our Talent Acquisition Manager, Mark Butler, writes this article for anyone interested in applying to Deepgram. He says, “I don’t write code. I’m not building models. My world is Talent: finding great people, helping them thrive, and shaping the culture that makes them want to stay. But every day, I rely on AI tools to do my job better, faster, and smarter. And we expect the same from our job candidates.”

Banking’s Next Front Door Is Voice: How CX Leaders Are Rewiring Service for Speed, Trust, and Scale - Banking has always been about trust, but customer expectations are shifting fast. The two key tradeoffs are speed versus security and efficiency versus empathy. Learn how AI can solve these problems in this article!

🤖 Bonus Bits and Bytes!

• 🔊 Google launches “Search Live”: real-time AI voice + camera search - Google rolled out a new feature called Search Live in the U.S., letting users engage with search via voice and visuals in real time.

• 👂 Study: AI-generated voices now “indistinguishable” from real - A recent study compared human voices vs synthetic ones (some cloned, some from general voice models) and found that in many cases listeners couldn’t tell the difference.

• 📊 ElevenLabs valuation & internal share sale - ElevenLabs (voice cloning / synthesis firm) is offering a tender for employees to sell shares, valuing the company at ≈ $6.6 billion — roughly double from earlier in 2025.