Modulate

Modulate is now #1 for Mandarin UER on Sierra's μ-Bench 🥇 Sierra's μ-Bench is the multilingual ASR benchmark built from real customer service calls. On it, Modulate Transcribe beat Google Chirp-3 and ElevenLabs Scribe-v2 on Utterance Error Rate - in Mandarin, the hardest locale on the leaderboard. 27.4% UER. 658ms P95 latency → nearly 2x faster than Chirp-3. The lowest reported Mandarin UER on the leaderboard, at nearly half the latency. Test it on your own audio. 1,000 free credits: bit.ly/4wqTg1k

We've barely scratched the surface of what voice AI can understand. Transcription was the first chapter - #1 out of 88 on @HF ASR/STT open leaderboard 💪🏅 Understanding behavior is the next That's the direction we're building toward at Modulate: helping organizations detect, discover, and act on the signals hidden inside conversations. Learn more at bit.ly/4fdgn7T

Heading to @Ai4Conferences next week? Join us and @axonis for happy hour on Tuesday, August 4 from 6-8 PM at Sugarcane, The Venetian. Good drinks, great conversations, and even better company. Whether you're building AI, deploying it, or just want to meet fellow practitioners, we'd love to see you there. Register to join here: lnkd.in/eDZT8VVC See you in Vegas! 🎲

Everyone says their Voice AI is "state of the art" 👀 Ask better questions. Our Voice AI Buyer's Guide covers the evaluation criteria that actually matter - from latency and reliability to security, scalability, and total cost. Because the best demo isn't always the best product 🚨 Download it here: bit.ly/4yrX8Ay

When people think about AI-generated voices, they tend to ask one question: "Can it perfectly imitate a human voice?" We think that's the wrong question. The more interesting question is whether AI can faithfully recreate everything that surrounds a voice - the subtle timing, conversational flow, and behavioral patterns that make real interactions feel... real. That's where the conversation gets much more interesting. If you're interested in how we're approaching voice authenticity and deepfake detection, explore what we're building at bit.ly/3R9whbE.

View the leaderboard here: research.sierra.ai/mubench/

Modulate is now #1 for Mandarin UER on Sierra's μ-Bench 🥇 Sierra's μ-Bench is the multilingual ASR benchmark built from real customer service calls. On it, Modulate Transcribe beat Google Chirp-3 and ElevenLabs Scribe-v2 on Utterance Error Rate - in Mandarin, the hardest locale on the leaderboard. 27.4% UER. 658ms P95 latency → nearly 2x faster than Chirp-3. The lowest reported Mandarin UER on the leaderboard, at nearly half the latency. Test it on your own audio. 1,000 free credits: bit.ly/4wqTg1k

Last week, the Modulate team had a great time at Develop: Brighton. 🎮 It was an incredible few days of reconnecting with partners, customers, and friends across the industry, meeting new faces, and having thoughtful conversations about the future of voice AI in games. One of the highlights of the week was co-hosting a dinner with @checkstep, bringing together great people from across the games industry for an evening of conversations, good food, and even better company. Thanks to everyone who stopped by to chat and joined us throughout the week. See you next year! 👋

Building great technology starts with building a great team. This week, we took a break from our desks to celebrate summer with the people behind @modulate_ai. Great food, great conversations, and plenty of laughs made for a memorable afternoon together. We're grateful for the team that makes Modulate what it is - and excited for everything we're building together next! ☀️

Scammers don't need to fool you for an hour. They just need one moment where urgency wins over skepticism. As AI-generated voices become more convincing, the safest response isn't trying to decide whether the voice is real. It's verifying the person through a completely different channel 💡 Simple habits beat sophisticated scams 🤷‍♀️ What's your go-to way of verifying an unexpected or urgent request?

Our Voice AI Buyer's Guide breaks down the evaluation framework we wish more teams used - from latency and reliability to deployment, security, pricing, and the hidden tradeoffs that don't show up in benchmark charts 👀🚨 If you're evaluating Voice AI vendors in 2026, start here. Download the guide: bit.ly/3QZAdvy

@SeharShinwari @hackernoon couldn't agree more. we built our models with the same mession - and are adapting on nearly a daily basis

@hackernoon @modulate_ai AI-generated music is getting too good to easily spot—detection tools need to evolve fast or this will only get messier.

AI music is flooding streaming platforms faster than reviewers can handle, and standard detection tools can't keep up with real-world catalogs. @modulate_ai explains how their API tackles AI music detection at scale: hackernoon.com/detecting-ai-g…

@sophiep33_l @huggingface why not test it out for yourself to find out? we are certain we'll succeed.

@modulate_ai @huggingface 1 out of 84 is cute but can it understand me after 3 drinks

Modulate Transcribe just ranked 1 out of 84 models on @huggingface Open ASR Leaderboard. Beat ElevenLabs. Beat AssemblyAI. Beat Deepgram. On the industry's MOST trusted transcription benchmark. 🧵

Most companies analyze conversations to answer one question: 👉 "What went wrong?" We think they're asking too little. Voice conversations can reveal customer experience, agent coaching opportunities, product feedback, emerging risks, and much more. @mpappas74, our CEO explains why that's the future of voice AI.

We put together a practical guide to help technical and business teams evaluate Voice AI beyond the marketing claims. Get your copy of the playbook here: modulate.ai/lp/voice-ai-bu…

Choosing a Voice AI platform based on benchmark numbers alone is like hiring an engineer based on typing speed 👀 The real questions are: ⁉️ How does it perform in noisy, real-world audio? ⁉️ What happens when speakers overlap? ⁉️ Can it scale? ⁉️ What's the true cost of production?

Soon, every platform that hosts audio will need to answer one question: Is this AI-generated? We teamed up with @hackernoon to show developers how to detect AI-generated music using the Modulate API. Read the walkthrough and start building: bit.ly/4vqr2lJ

A little Modulate tradition we look forward to every week! Last week's Social Night brought the team out into the beautiful streets of Somerville to unwind, connect, and enjoy a great evening together 💜

Modulate Transcribe just ranked #1 out of 84 models on @huggingface Open ASR Leaderboard. Trained on 500M+ hours of real-world, noisy audio - and still costs $0.03/hour. Incredibly proud of my team at @modulate_ai! We're outperforming players with 10x our resources, on a benchmark anyone can go check for themselves. Nothing beats building the best product and letting the numbers speak. Check out the leaderboard here: bit.ly/4b6yWK3

View the leaderboard here: huggingface.co/spaces/hf-audi…

Modulate Transcribe is trained on 500M+ hours of real-world, noisy audio - not clean studio clips. That's what shows up in production, not just on a leaderboard. Don't take our word for it. Test it yourself. 1,000 free credits on us: modulate.ai/api/speech-to-…