Shillong, Oct 15: Shillong-based AI research lab MWire Labs showcased its Garo speech-to-text model at Google’s Lab to Impact: A Dialogue on India’s AI-Powered Future, held at The Imperial in New Delhi on October 12.
The event brought together researchers, startups, policymakers and industry leaders to explore how artificial intelligence research can be translated into real-world applications across India.
MWire Labs demonstrated its model live before policymakers, industry leaders and the media, marking a significant step towards bringing indigenous languages into mainstream voice AI systems. The model is the first publicly available speech-to-text system for Garo, a language spoken by more than a million people across Meghalaya and Assam that has largely remained underrepresented in mainstream AI technology.
The open-source model was trained and evaluated using conversational speech data from Project Vaani, a large-scale speech collection initiative led by the Indian Institute of Science (IISc) and ARTPARK, with support from Google.
According to results published by MWire Labs, the model achieved a 97.5 per cent relative improvement in character error rate compared with the zero-shot baseline, highlighting the gains made through training on language-specific speech data.
“Languages of the Northeast are often missing from the AI systems people use every day,” said Badal Nyalang, CEO of MWire Labs.
“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,” he added.
The Garo model forms part of MWire Labs’ broader effort to develop foundational language AI technologies for more than 220 indigenous languages of the Northeast. Its work spans speech-to-text, text-to-speech and translation, with the aim of making digital technology more accessible to communities in their own languages.
The lab’s Lemka voice stack already supports six regional languages: Khasi, Garo, Mizo, Meitei, Nagamese and Kokborok.
Through these initiatives, MWire Labs aims to bridge the gap between indigenous languages and emerging AI technologies, enabling people across the Northeast to interact with digital platforms in the languages they speak at home.
