Krisp

Last week in Voice AI 🔥 Open speech models and on-device generation are expanding who can build voice AI, while sales platforms and phone systems are expanding how it reaches customers. Here's what stood out 👇 - Block unveiled Dessa, a full-duplex speech model trained only on audio that Jack Dorsey calls the most authentic yet. - ElevenLabs launched Eleven Realtime, a speech-to-speech model family that starts replying in about half a second. - IBM released Granite Speech 5.0 as open-weight ASR under Apache 2.0, covering six languages with translation to English. - Ant Group open-sourced Ming-3.0, five MIT-licensed speech models spanning TTS, audio understanding and recognition. - Gradium open-sourced Phonon, a 31M-parameter on-device TTS model that starts speaking in about 7ms on a laptop CPU. - Scaled Cognition launched Voice, a low-cost enterprise TTS model at $9 per million characters and 70ms latency. - Krisp is a finalist for Most Innovative CX Technology in CX Today’s 2026 CX Awards. - Amazon Web Services (AWS) upgraded the Polly generative engine, adding 14 voices plus Hindi and Finnish. - OpenAI’s GPT-Live added prompt-to-voice, letting developers design a custom voice from a text description in one call. - monday.com launched Harmony, a sales voice-agent platform using task-specific models at about 200ms latency. - Synthflow AI went live on Zoom Phone and Contact Center, with Batteries Plus deploying it across 700+ US stores. - Deutsche Telekom tapped AudioCodes to add AI call summaries, powered by Mistral Small 4, to business mobile plans. - Samsung Electronics added Malay to Galaxy AI, bringing Live Translate on calls to 24 languages. - VoxNexus launched on-device voice AI for Android, claiming always-on listening at about 2.5% of battery over 12 hours. - Nirva raised $25M for an AI pendant that journals the wearer’s day and is built to capture only their voice. - ZDNET cloned a voice with free tools and ran a scam call on family, showing how accessible voice fraud has become. - Phonely made the case for measuring voice AI by revenue, citing customer bookings and conversions over cost savings. - Voqalize argued voice is becoming the default software interface, pitching a voice layer that drops into existing apps. Image source: ElevenLabs

Your customer needs help. Your agent knows the answer. Why wait for an interpreter? With Krisp Voice Translation, agents and customers each speak their own language while AI translates both sides of the call live. Krisp translates. Your agents help. Watch it in action, then try it for free.

Your agents already know how to help. Imagine what they could do in 65+ languages. Krisp Voice Translation translates customer calls in real time, helping your existing team handle more multilingual conversations and reduce interpreter spend. Hear it in action. Try Krisp for free.

Krisp just ranked #2 out of 500 AI companies for trustworthiness. 🏆 Cybernews’ 2026 AI Trustworthiness Ranking looked at 500 AI companies. Krisp came in #2, with perfect scores in Security, Data Privacy, and Organizational Transparency. Big win for the team. But the engineer in me is excited about why we scored this way. At Krisp, we think hard about what AI should do. We think just as hard about what it shouldn’t do: - It shouldn’t collect more data than it needs. - It shouldn’t move your data around just because it can. - It shouldn’t make you guess what happens to your voice. Here’s the secret: most good security decisions are just good engineering decisions in disguise. Process on the device when you can. Move less. Store less. Keep the system simple enough to explain in one breath. Data you never store can’t leak. It also needs no backups, no migration plan, and no 2am incident call. Everybody wins. Especially the person on call. Huge thanks to the team for building products that hold up when someone looks really, really closely. Turns out boring architecture is pretty exciting. 😉 Full deep dive on how we approach security in the comments 👇

Last week in Voice AI 🔥 Voice AI is moving faster than enterprise adoption, and the race to automate conversations is raising new questions about trust, privacy, and security. Here's what stood out 👇 - Amazon ships Nova 2.5 Sonic, a speech-to-speech model on Bedrock with stronger reasoning, tool calling, and lower latency for real-time voice agents. - Sierra launches Fleming-1, a model that scores a caller’s audio in real time to flag whether the voice on the line is an AI agent. - Krisp makes its flagship Voice Isolation model 2x faster with no quality loss. - Automation Anywhere agrees to acquire boost.ai, moving into customer-facing conversational and voice AI. - Apple acquires Huxe, a personalized-audio startup from ex-NotebookLM developers. - Sarvam launches Kivi for Mac, a voice productivity app powered by Saaras V4 that dictates and edits across apps in 22+ Indian languages. - Krisp ranks #2 of 500 on Cybernews’ AI Trustworthiness index, scoring 97/100 with perfect marks on security, transparency, and privacy. - Google opens its SynthID detector to the public, a website that checks whether media carries Google’s AI watermark. - Google releases AI Edge Foresight, a free offline Mac note-taker that transcribes and answers questions locally. - Google’s Pixel 11 adds Call for Me, letting Gemini place routine business calls on your behalf with a live transcript. - DeepL Voice makes Fast Company’s Next Big Things in Tech 2026, cited for voice-to-voice translation that preserves each speaker. - A nasscom-Krisp report finds only 13% of Indian enterprises have scaled voice AI, with 31% in limited production. - eHealth, Inc. moves its voice AI from pilot to full-time, using REGAL’s Alice to prescreen up to half a million Medicare calls per enrollment window. - EliseAI partners with Nextech Systems to co-develop agentic voice AI for ophthalmology, dermatology, and other specialty practices. - AudioShake launches The Refinery, turning overlapping real conversations into speaker-separated AI training data. - Rebel Audio raises a $10M Series A for an AI podcast production platform with translation, clipping, and one-click publishing. - Sonilo raises $11M, led by B Capital, for a model that generates music and sound effects synced to video. - AncSonic debuts an on-device audio AI architecture, its SORC design targeting earbuds and headphones on low-memory silicon. - Voice AI lawsuits put contact centers on BIPA watch, as new voiceprint suits against Walmart and Lowe’s extend the theory to customer service. - OpenAI plans to retire its legacy TTS voices, drawing pushback from creators who built around tts-1 and gpt-4o-mini-tts. - The Verge argues we can’t quietly redefine “recording”, as always-on AI hardware blurs what counts as captured.

Voice AI is getting closer to a point where the technology is no longer the biggest constraint. I sat down with Sam Krut, Co-Founder and President and of Flip, to talk about what it takes to build voice agents that can automate most of the contact center, from model architecture and system access to workflow depth, business processes, and where humans still fit. Here’s what stood out most: - Vertical depth is becoming a real moat because repeatable workflows and integrations matter more than a generic agent that can talk about anything. - An agent that can’t access systems and take action is still just a better IVR. - 70–90% automation flips the contact center model from humans supported by AI to AI handling the work and humans taking the exceptions. - Better models won’t unlock higher automation if the business still runs on weak APIs, fragmented systems, and manual processes. - The hardest knowledge problem isn’t retrieval. It’s turning human judgment, exceptions, and workarounds into something machines can reliably execute. - Containment is a weak success metric if the customer still has to call back. - The future stack may use smaller, faster models for real-time tasks and larger models for context, reasoning, and verification. - Speech-to-speech is still easier to demo than deploy when context and tool use break down in longer conversations. - Voice AI may now be closer to technical readiness than organizational readiness, with enterprises increasingly setting their own ceiling on automation. - BPOs face a business-model shift more than an extinction event, as value moves toward process expertise, exception handling, and AI-enabled operations. The question now is how much of the work behind that conversation it can actually own. Full ep and takeaways: https://lnkd.in/gaTgsd6Q