Bengaluru: Chariot, an AI research lab set up by former researchers from Google DeepMind and Meta's FAIR lab, on Thursday launched Chariot-1-ASR Realtime, a live speech recognition model. The company says it recorded fewer errors than every rival tested across 10 of India's most spoken languages.
The model is one of the first four teams selected under the IndiaAI Mission. Most speech recognition models are built for a single task, while Chariot-1 is a foundation model for speech, the company said. It learns speech the way large language models learn text.
Speech recognition comes in two kinds. Live systems, used in voice agents, call centres and captions, write text as someone speaks, while batch systems wait for a recording to finish. Chariot says its live system beat the best of both.
Across 79 public test sets, the company said the model made 9 per cent fewer errors than the strongest batch system and 26 per cent fewer than Google's Gemini 3.5 Transcribe. In Hindi it made 23 per cent fewer errors than the next best. Chariot also claimed top spot on real Hindi and Indian-English phone calls, on district-level Hindi accents, and on noisy recordings.
The model handles regional accents and code-mixing, and can write Hindi in Devanagari or in Latin script. Banks, insurers, telecom operators and public helplines can serve customers across 10 languages with one model, the company said. The model is available on platform.chariot.in. The company noted that the results have not been independently verified. The company said further releases are expected across speech recognition and speech generation.