Events
DeepLearning.AI Voice AI Hackathon: In-person hackathon with Sabre + Vocal Bridge. Teams build voice agents that book real trips end-to-end, judged by Andrew Ng. (Jul 18, Mountain View, Luma)
Voice Agent Evalathon: Okareo x Telnyx virtual challenge to red-team a voice agent’s reasoning, execution, and stability. (Jul 15, Virtual, Meetup)
Top Updates 💪
OpenAI launches GPT-Live, a full-duplex voice model that listens and speaks simultaneously, powering a more natural ChatGPT Voice experience. (OpenAI)
xAI releases 21 multilingual flagship voices for Grok, expanding its voice lineup across multiple languages. (xAI)
Gradium raises $100M seed backed by NVIDIA, making the Paris-based Kyutai spinout one of the largest seed rounds ever for ultra-low latency voice AI. (TechCrunch)
Rylo AI raises $85M to scale its real-time captioning platform for deaf and hard-of-hearing users, rebranded from Nagish. (AlleyWatch)
Five9 launches next-gen Voice AI Agents built on a purpose-built agentic architecture (Insider Monkey)
Taco Bell expands voice AI to 890+ drive-thrus across 38 states, powered by Omilia. (FER Magazine)
Omilia launches Lexis, a native generative TTS engine for enterprise CX with sub-45ms latency. (AIthority)
Dell AI Factory partners with Deepgram and Penguin Solutions to deliver enterprise-grade real-time voice AI infrastructure. (Business Insider)
SoundHound’s OASYS platform wins “Agentic AI Company of the Year” as Q1 revenue hits $44M, up 52% year-over-year. (Nasdaq)
CallTower partners with Sestek to add conversational AI, voice biometrics, and real-time translation to its CX portfolio. (Telecom Reseller)
Whispp raises $5M to scale its on-device AI that reconstructs speech for people with voice disorders in real time. (Pulse 2.0)
AI voice agents boosted specialty care enrollment 340% in a peer-reviewed clinical study by RadiantGraph. (PR Newswire)
A $25M deepfake scam at Arup used AI-generated executives on a video call, becoming a landmark case for corporate voice fraud. (FinanceFeeds)
Reality Defender warns that autonomous AI callers can now bypass contact center defenses at scale, posing a new voice fraud threat. (Biometric Update)
Hydaway launches RealityChek, a streaming audio deepfake detector for enterprises that flags synthetic speech in real time. (Biometric Update)
Engineering Corner 😎
OpenAI ships GPT-Realtime-2.1-mini with reasoning and tool use support at 6x lower cost and 25% reduced latency. (Marktechpost)
AssemblyAI details how Universal-3.5 Pro handles noisy audio, sharing techniques for improving transcription accuracy on hard recordings. (AssemblyAI)
Flock Safety explains its audio detection system that identifies gunshots and crashes in real time using acoustic sensors and AI classification. (Flock Safety)
The most interesting part of full-duplex voice AI isn’t that conversation feels more natural.
It’s that AI can begin responding to reasoning as it unfolds.
In education, that could matter enormously. A written answer shows the final output. A live conversation can reveal hesitation, borrowed language, misconceptions, and the exact moment a student’s reasoning breaks down.
The opportunity isn’t simply an AI that talks more smoothly. It’s an AI that listens closely enough to ask the next precise question.