E-Paper
TwinMind Sets New World Record in Speech Recognition—Powering 140+ Languages for India and the World
TwinMind launches Ear–3 speech recognition model, enhancing transcription accuracy and local language support across 140 languages.
Menlo Park, California, September 12, 2025 — The global AI landscape has taken a major leap forward, and India is set to benefit in a big way. TwinMind, a technology innovator founded by three former Google X AI scientists, has unveiled its new Ear–3 speech-recognition model, setting new records in speech-to-text accuracy, local-language support, and affordability. Check here: twinmind.com/transcribe
A Game-Changer for India’s Diversity
With support for over 140 languages—including Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, and many more—Ear–3 marks a milestone for enterprises, creators, and everyday users across India. For millions who speak in their own dialects or mix languages in daily life, TwinMind now offers one of the world’s most accurate and usable transcription solutions.
- Record Accuracy: Just 5.26% Word Error Rate (WER)—the best score ever, surpassing previous leaders like OpenAI Whisper and ElevenLabs.
- Speaker Diarization: 3.8% Diarization Error Rate (DER), enabling remarkably accurate identification of individual speakers, even in group settings like team meetings or family chats.
- Lowest Cost: At just ₹19 per hour ($0.23/hour), Ear–3 is the most affordable among major AI speech-tech players.
- Comprehensive Language Coverage: Supports 40+ more languages than competitors—built with true “Indian accent” strength across regions.
“We started out building an AI assistant—something like ‘Jarvis’—for daily life,” said Daniel George, PhD, CEO of TwinMind, “and we ended up inventing the world’s best speech recognition technology by accident.”
“Speaker separation in Indian group conversations—where people freely mix English, Hindi, and local languages—was a huge challenge,” added Sunny Tang, PhD, CTO. “Ear–3 solves this pain point.”
Mahi Karim, PhD, co-founder: “For my mom, who switches seamlessly between Hindi, Bengali, and English in one sentence, Ear–3 finally gets it right. Others can’t keep up.”
For press or demo requests:
Jerome Manuel, Head of Marketing
jerome@twinmind.com
About TwinMind
TwinMind is a personal AI app that listens continuously and works as an extension of human memory and intelligence. Headquartered in Menlo Park, California, TwinMind was founded in 2024 by three former Google X AI scientists.
The TwinMind mobile app pairs with a Chrome extension to integrate memory from what you see on browser tabs (Email, Slack, Notion, Zoom, etc.). TwinMind’s iOS app launched earlier this year and is already used by tens of thousands of people to capture hundreds of thousands of meetings, lectures, and conversations. Recently, TwinMind publicly launched on Android and received 10,000+ downloads in its first week.
Funding: TwinMind has raised $5.7M at a $60M valuation. In the latest round, the company raised funding from Stephen Wolfram (the only startup he’s invested in in 30+ years), Streamlined Ventures, Sequoia, and others. Previous backers include Anand Rajaraman (earliest investor in Facebook and founder of two startups with $250M+ exits each), Dan Roth (Oracle Head of AI; Distinguished Professor at UPenn), and Michael Liou (first check in Robinhood and Zapier).
Team: All three co-founders are former Google X scientists (PhDs in AI). The three additional full-time founding members in the Bay Area bring marketing, design, and operations experience scaling B2C startups to 3M+ users and $75M+ ARR.
Tech: TwinMind recently set records for both transcription accuracy and speaker-diarization accuracy (across 140+ languages). It is also the only mobile app enabling 12+ hours of continuous transcription without draining the battery. TwinMind’s on-device approach seamlessly switches from edge to cloud, lowering costs by 20× while preserving full user privacy. Its “memory moat” (similar in spirit to Google Photos) compounds over time, driving higher retention than typical AI chat apps.
Competitive Landscape
Made for Enterprises, Everyday Use, and India’s Tech Ecosystem
- Perfect for Indian corporates: Accurate meeting notes, hiring, compliance, sales calls, and customer service—across Indian languages and accents.
- Ideal for creators and educators: Effortless content production, subtitles, podcast transcripts, and more.
- Privacy-first, built for India: Runs on-device to keep data safe and transcribes for 12+ hours on phones without draining battery. From Bengaluru, Mumbai, and Delhi to rural India—Ear–3 adapts to low bandwidth and diverse backgrounds.
How to Get Started - The TwinMind web platform is live for instant audio transcription and summaries.
- Indian developers and businesses can join the API waitlist for early access.
- Android app for India is now available—bringing world-class AI to every region, including beyond large metro cities.
TwinMind Features
Beyond record-setting accuracy, TwinMind offers 12+ hours of continuous transcription without battery drain, seamless speaker diarization for meetings, and support for 140+ languages and accents—making it a truly global speech model. With its web platform, iOS app, and Android app, TwinMind integrates easily into daily workflows. Learn more at www.twinmind.com.
Transforming the Way India Works and Communicates
Whether for businesses, content creators, educators, or anyone capturing ideas across India’s linguistic richness, TwinMind Ear–3 is set to transform how we interact with technology.
Note to readers: This article is part of HT's paid consumer connect initiative and is independently created by the brand. HT assumes no editorial responsibility for the content, including its accuracy, completeness, or any errors or omissions. Readers are advised to verify all information independently.
Want to get your story featured as above? click here!
