New Delhi, November 12, 2026: Chariot, a frontier AI research lab founded by former researchers from Google DeepMind and Meta’s FAIR lab, and one of the first four teams selected under the IndiaAI Mission, today launched Chariot-1-TTS-Multilingual, its new multilingual text-to-speech model.
The model brings together 11 Indic languages and English, more than 50 voices, and code-mixing between any two supported languages within the same sentence. It can clone a voice from just a few seconds of reference audio and generate speech from up to 1,000 characters in a single request.
Turning text into speech involves understanding how it should be said. A payment amount needs the right spoken value and currency. “B+” needs a different reading when it refers to a blood group and when it describes a grade. Dates, abbreviations and technical expressions bring similar choices. Chariot-1-TTS-Multilingual uses context to work out how written expressions should be spoken, with this reasoning built into the model to reduce the need for custom normalization rules or carefully engineered LLM prompts.
Chariot evaluates speech generation on three tiers: intelligibility, reasoning accuracy and quality.
The accuracy evaluation includes demanding speech-generation stress tests. Alongside everyday and conversational text, the intelligibility suites test code-mixed sentences and numerical and technical expressions. The verbalization benchmark targets the contextual decisions involved in turning written numbers, symbols and codes into speech.
Its evaluation compares selected model versions from ElevenLabs, Cartesia, Microsoft, OpenAI, AWS and Sarvam, with coverage varying by benchmark. Chariot-1-TTS-Multilingual ranked first among the configurations reported for each of the following measures:
- Intelligibility: The lowest character error rate, at 7.60% on Hindi and 6.47% across nine other Indic languages, across general text, numerical and technical robustness tests, and native conversational text. The nine-language mean gives equal weight to each language and test set.
- Reasoning accuracy: The highest English verbalization accuracy at 93.78%, on a challenging benchmark of 1000+ sentences across 40+ categories, including currency amounts, dates, ordinals, URLs, chemical formulas and stock tickers.
- Quality: The highest plain-English NISQA score, an automated prediction of speech quality, on a benchmark of English texts.
Parth Sarthi, Co-Founder, Chariot, said: “People use numbers, symbols and abbreviations, and move between languages in the same sentence. A speech model has to understand what those choices mean and how to say them. With Chariot-1-TTS-Multilingual, we have built that understanding into the model, so developers can spend less time writing normalization rules and tuning prompts, and more time building voice experiences that communicate what they intend.”
For businesses, the model supports voice agents, customer-service systems, learning applications and multilingual content. A bank can use it for payment reminders, a learning platform for narration, and a public helpline for spoken information across languages. Developers can choose from the voice catalogue or create a voice from a short reference recording.
The release extends the Chariot-1 model family’s work across speech recognition and speech generation. Chariot’s broader approach treats understanding and speaking as connected capabilities, with a research focus on models that understand, generate and reason in speech. The multilingual TTS launch brings that direction into applications that need to speak back to users.
Suvrat Bhooshan, Co-Founder, Chariot, said: “For voice technology to be useful in India, it needs to work with the languages people speak and the way they move between them. Our ambition is to build foundation models that understand, generate and reason in speech. Our work on speech recognition and speech generation takes that ambition into products developers can use, while building frontier AI from India for users here and around the world.”
Learn more about Chariot at chariot.in and explore the platform at platform.chariot.in. Full benchmark results, compared model versions and methodology: Introducing Chariot-1-TTS-Multilingual.
