1/ A big week for Voice AI.
A near-trillion dollar AI funding round, major platform partnerships, translation models rapid expansion, and new research exposing security risks in voice systems.
Here's what stood out 🧵
7/ A new study shows inaudible audio commands can hijack AI voice models unheard by humans via @jasonnelson for @DecryptMedia
Voice AI adoption will only grow as fast as trust. Security and adversarial resilience may end up being just as important as latency, accuracy, or naturalness.
https://t.co/gs86DbMOP4
Your meetings are now searchable inside @ChatGPTapp.
We just dropped on ChatGPT Apps. Connect your Krisp account in seconds and ask: who said what, what's pending, what did we decide.
No more digging through notes. Just ask.
We just shipped one of our most requested features 🚀
Global AI Chat.
Until now, AI Chat worked per meeting. Now it works across all of them.
No more digging through calls or piecing things together.
One question. One answer. Across everything.
6/ Voice AI is getting smarter, but the companies that win will solve the messy physics of real conversations, not just model benchmarks.
Full digest: https://t.co/5EVlMSAVP6
5/ @WisprFlow bets on India as its fastest-growing market with Hinglish dictation support, 2.5M downloads, and 100% month-over-month growth via @JagmeetS13 for @TechCrunch
India may be the most important proving ground for voice AI. Multilingual switching, accents, noisy environments, mobile-first behavior. If your system performs there, it’s likely robust anywhere.
https://t.co/TYmtmfVIje
Already powering Daily, Vapi, LiveKit, Vodex, Ultravox and the world's largest AI labs.
The next generation of voice agents won't just talk. They'll listen.
Available now → https://t.co/YVHfXxfonz
We just shipped VIVA 2.0 at @krispHQ — launching live at @twilio Signal in SF.
8 years. A trillion minutes of real-world voice. This is what it all built toward.
Voice Infrastructure for Voice AI Agents 🧵
Every voice agent builder hits the same wall: demo works, production doesn't.
VIVA 2.0 — one SDK, sits before STT:
- Voice Isolation v3: isolates the primary speaker's voice, improves WER
- Turn Prediction v3: predicts end-of-turn from audio, 13+ languages
- Interruption Prediction v1: first model to tell "uh-huh" from "wait, stop"
- Signal Detectors: an all new category of perceptual models that classify synthetic speech, gender, and accent in real time
We won 🚀
Krisp is a 2x Webby Winner for Technical Achievement for our Voice AI:
🏆 Webby Winner (@iawchs)
📢 People’s Voice Winner
13,000+ entries. 70+ countries. 4.6M votes.
This one means a lot. Thank you for the support!