Chariot Launches Chariot-1-TTS-Multilingual Speech Model
Chariot released Chariot-1-TTS-Multilingual on November 12, 2026, with support for English and 11 Indic languages, according to the company. The text-to-speech model offers more than 50 voice choices and can produce speech from sentences that mix two supported languages, Chariot said.
The company’s language list includes Hindi, Urdu, Punjabi, Bengali, Gujarati, Marathi, Odia, Kannada, Telugu, Tamil and Malayalam, alongside English. Chariot said each generation can handle as many as 1,000 characters, while voice cloning requires a reference recording lasting a few seconds.
For written material with more than one possible spoken interpretation, Chariot said the model uses context to select a reading. This covers amounts, abbreviations, dates and symbols. One company example uses B+ in two settings: a school grade and a blood-group result. According to Chariot, the model handles these distinctions internally, reducing the need for separate rules or prompts to prepare text for speech.
Chariot’s evaluation, dated November 11, 2026, covered 8,655 configured text cases in English and ten Indic languages. Comparisons applied only to the configurations included in each benchmark. For intelligibility, the company compared transcripts produced by automatic speech recognition with normalized reference text. Its separate English audio-quality assessment used NISQA, an automated predictor of speech quality.
On the English interpretation benchmark, Chariot reported accuracy of 93.78% across 1,029 sentences in 42 categories. Those categories included chemical formulas, currencies, codes, dates, stock tickers, ordinals and URLs. The company defined accuracy as the proportion of assessed outputs whose spoken interpretation was acceptable, and placed the model first among the configurations tested in that benchmark.
For Hindi, the company reported a character error rate of 7.60% from 644 prompts. Across nine other Indic languages, its reported average was 6.47%. That group comprised Tamil, Telugu, Punjabi, Odia, Marathi, Malayalam, Kannada, Gujarati and Bengali. Chariot said the tests covered conversational text as well as general, numerical and technical robustness, with equal weighting for every language and test set in the nine-language average. The company placed the model first among the compared configurations on both measures.
In the separate English audio-quality comparison, Chariot reported a NISQA score of 4.901 from 1,088 prompts written in plain English. The company ranked the model first within that test’s set of configurations. Unlike the intelligibility assessment based on speech-recognition transcripts, this result was an automated estimate of audio quality.
* This article is based on publicly available information at the time of writing.