Shillong, Oct 14: A Shillong-built AI that understands Garo took centre stage at Google’s national AI showcase.
MWire Labs, a Shillong-based AI research lab, showcased the first publicly available Garo speech-to-text model at Google’s Lab to Impact: A Dialogue on India’s AI-Powered Future held in New Delhi on October 12.
The Google event brought together top researchers, startups, policymakers and industry leaders to show how lab research has a real-world impact.
MWire Labs did it live. The team demonstrated the Garo model to policymakers, industry leaders and media, turning spoken Garo into text in real time.
Garo is spoken by over a million people in Meghalaya and Assam, but has been almost invisible in mainstream voice AI.
According to the lab, it was built on conversational speech data from Project Vaani, the Indian language speech collection initiative by IISc and ARTPARK, supported by Google. MWire says it achieved a 97.5 per cent relative improvement in character error rate over the zero-shot baseline.
“Languages of the Northeast are often missing from the AI systems people use every day. Project Vaani gave us the speech data. Our team in Shillong built the models. Seeing Garo demonstrated live on this stage shows that homegrown work from the region can stand at the national level,” CEO of MWire Labs Badal Nyalang said.
MWire Labs is building foundational language AI for the Northeast’s 220+ indigenous languages — speech-to-text, text-to-speech, and translation. Its Lemka voice stack already covers six languages, including Khasi, Garo, Mizo, Meitei, Nagamese and Kokborok.
