In a major breakthrough for the global AI landscape, a new speech-recognition model named Ear–3 has been unveiled by TwinMind, a technology innovator founded by three former Google X AI scientists. The new model sets unprecedented records in accuracy, local-language support, and affordability, with a particularly significant impact expected in India.
Unveiled on Friday, the Ear–3 model has achieved the industry’s lowest Word Error Rate (WER) of 5.26%, a new global benchmark that outperforms leading services from competitors. Beyond accuracy, the model also boasts a remarkably low Speaker Diarization Error Rate (DER) of 3.8%, making it exceptionally adept at identifying and separating multiple speakers in a single conversation.
The most transformative aspect of Ear–3 is its unparalleled linguistic capability. With support for over 140 languages, it is the first and only model to provide true global coverage. This includes flawless handling of multilingual and mixed-script speech, which is a game-changer for linguistically diverse countries like India. The model’s creators highlight its robust performance in handling regional dialects and code-switching—the seamless blending of multiple languages within a single sentence.
According to Mahi Karim, a co-founder of TwinMind, “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.” This capability is set to revolutionize everything from business meetings and call centers to content creation and educational tools across India.
At just $0.23 per hour, Ear–3 is also the most affordable among major AI speech-tech players, making enterprise-grade quality accessible at a large scale. TwinMind’s on-device processing approach, which allows for continuous transcription for over 12 hours without significant battery drain, further enhances privacy and efficiency.
The company, founded by a team of former Google X scientists with PhDs in AI, has already secured significant funding from influential investors, highlighting the market’s confidence in its technology. With its record-breaking performance and focus on accessibility, Ear–3 is poised to democratize advanced speech AI, with India set to be a major beneficiary of its capabilities.
