ai-coustics

Building a Voice AI application? You can test our models for free in the Developer Platform: 🔗bit.ly/3Pn0KSB Still not sure which model is right for your use case? Let us know here👇

We’ll keep saying it: what sounds good to people doesn’t always work well for machines. After years of working on real-time speech, we know different use cases need different treatment. We built two focused model families: 🐦 Quail for Voice AI 🐦 Rook for human listening 👇

Curious how to get started with our SDK? Our DevRel Engineer, Pawel (@GrungeCoder) takes you through everything new in the ai-coustics Developer Platform - a faster path from sign-up to clean audio in your stack. Try it out: bit.ly/4d3OuhL

🎬Nest Talks #3: From waves on the shore to rally cars in the forest. Our newest team members Nader Soliman, Anil Keshwani, and Pawel Lach share what got them into audio and AI, and their personal relationship with sound.

🐦2.1 S: 10x smaller than 2.0, still outperforms it. Built for high call volumes and edge deployments. 🔷2.1 L: best-in-class quality, 25% lower compute. More concurrent calls, same resource cost. Try it for free in our developer platform now: bit.ly/4d1NJp9

We recently launched Quail Voice Focus 2.1 and our co-founder and Head of ML, Tim Janke, is here to show you what it can do. It’s our most lightweight model yet, with improved performance across the board.

Hear from our DevRel Engineer Pawel Lach (@GrungeCoder) as he takes you behind the scenes of Big Berlin Hack - where we hosted the Voice AI in the Wild track together with @tellidotcom, alongside partners like @GoogleDeepMind, @peec_ai, and @GradiumAI. Full recap below.

🐦 Voice Focus 2.1 S: 10x smaller than Voice Focus 2.0, still outperforms it. Built for high call volumes & edge deployments. 🔷 Voice Focus 2.1 L: best-in-class quality, 25% lower compute. More concurrent calls, same resource cost. Read more on our blog: bit.ly/4vNIGBu

What if you could get all the benefits of our flagship speaker isolation model, but 10x more efficient? Today we're launching Quail Voice Focus 2.1: a tenth of the compute of its predecessor, with better real-world performance across the board.

Nest Talks #3 🎬 Join us in welcoming three of our newest team members Nader Soliman, Anil Keshwani, and Pawel Lach. From expert bakers to origami connoisseurs, these three have hidden depths. Now they're building the future of voice AI.

👉Try it out in the Developer Platform: bit.ly/4mHnXLi

The days are getting lighter, and our Dev Platform is getting darker. 🧪 SDK Playground - test our Quail models right in the dashboard, no install needed. 📊 SDK Usage - minute consumption, per model, any time range. 🌙 Dark mode for long debugging sessions.

Evaluating Voice AI audio infrastructure shouldn’t take weeks. Voice Focus 2.0 is our newest speaker isolation model. Now you can evaluate it immediately, for free. Watch this guide to our @huggingface demo space, and test the model yourself in minutes. eu1.hubs.ly/H0tzv4Z0

🟢 Lower Word Error Rate (10-43%) 🟢 Fewer insertion errors 🟢 Consistent speaker clarity across major ASR providers Learn more on our blog: bit.ly/3PJT1xO

Three weeks ago we rolled out our most powerful speaker isolation model: Quail Voice Focus 2.0 It strips competing speakers, background noise, echo and codec artifacts before audio ever reaches your ASR. If you’re building voice agents, this is where accuracy is won👇

Read more on our blog: bit.ly/4v31JY5

Nest Talks #2: Voice agents often fail because they hear too much. Our Co-founder & Head of ML, Tim Janke shares the thinking behind Voice Focus 2.0, and why isolating the primary speaker in real-world audio is harder than it seems.

If you’re building speech systems, this is the dataset you’ll want to benchmark against. 🤗Explore our dataset: huggingface.co/datasets/ai-co…

Nest Talks #1: Dawn Chorus - introduced by our Head of Data, Joschka Wohlgemuth Foreground speaker isolation under real-world conditions is a major challenge for speech systems - and that’s exactly what our new open-source dataset on @huggingface is built to evaluate.

For developers this means: 🎯 No background voice interference 🗣️ Integrated VAD ⚡ Better ASR reliability Learn more on on our blog: bit.ly/47na3b2

The full ai-coustics audio intelligence stack is now built into @livekit - and it takes 3 minutes to integrate.