Text to speech for e-learning, courses and training modules

A lesson is dense with things a voice can get wrong: percentages, units, dates, abbreviations, amounts. Chariot reads written text the way a person says it. "500 mg", "Dr.", "06:45" and "₹1,20,000" come out as speech without any markup. In Chariot's own evaluation on English and Hindi, the model read hard written text (dates, currencies, URLs, units) correctly 94.0% of the time, first of the nine systems tested.

Clarity over a long module also depends on the voice sounding like a person, because learners stop listening to a flat one. Naturalness (NISQA) scored 4.91 in English and 4.97 in Hindi in the same evaluation, tied for first in English and first in Hindi.

The clip above is Alan, an American English voice, starting a lesson on fractions.

With a recorded narrator, a changed slide means a new session, a new invoice and, often, a slightly different sounding voice because months have passed. With Chariot the narration is the text. Change the sentence, generate again, and the new line is in the same voice at the same settings as the rest of the module.

This matters most for compliance training, product training and anything with a date or a price in it, where the content is right for a quarter and then not. It also means the narrator never leaves: a course made in one voice three years ago can be extended in that voice today.

Speed is set per request, from 0.5x to 1.5x, so a revision module can be read faster than a first-pass lesson without touching the script.

Chariot reads 15+ languages, including English, Hindi, Urdu, Bengali, Marathi, Telugu, Tamil, Gujarati, Kannada, Malayalam, Odia, Punjabi, German and Chinese. Each Indian language has native voices (12 for Hindi, 2 for each of the others), and every voice can read every supported language. A course can keep one narrator across all its language versions, or use a native voice per language.

Indian classrooms do not speak one language at a time. A Hindi lesson keeps the technical terms in English, and a Tamil one does the same. Chariot reads Hinglish and other mixes of an Indian language with English as one sentence, so "photosynthesis ka process" is read without a stumble at the switch. It also reads Hindi written in Roman letters, which is how many teachers write their notes.

In Chariot's own Hindi evaluation, the model had the lowest Hindi character error rate (7.73%) of the seven systems tested.

Yes. The API at POST https://api.chariot.in/v1/tts returns a WAV file for a block of text. For reading on tap, the streaming endpoint sends the first audio in about 75 ms (median, over WebSocket) and speaks sentence by sentence, so the learner hears the first sentence while the rest is still being generated. Each request takes up to 1,000 characters, which is a paragraph or a question at a time.

Speed from 0.5x to 1.5x is a request parameter, so you can give the learner a slider rather than choosing one speed for everyone. Younger learners and new readers usually want slower; revision wants faster.

Concurrency is set by plan: 3 requests at a time on Starter, 10 on Startup and 25 on Scale. There are Python examples to start from. Text and audio are processed in India.Concurrency is set by plan: 3 requests at a time on Starter, 10 on Startup and 25 on Scale. There are Python examples to start from.

If the instructor records 10 to 60 seconds of speech (about 30 seconds works best), the cloned voice is ready in about 10 to 20 seconds and speaks every language Chariot speaks. A teacher who recorded in English can narrate the Hindi version of the same course in their own voice.

Cloned voices are included on paid plans: 3 on Starter, 10 on Creator, 50 on Startup and 500 on Scale. Consent is required and you may only clone voices you have the right to use. For an institution, that means an agreement with each instructor whose voice is cloned.Cloned voices are included on paid plans: 3 on Starter, 50 on Startup and 500 on Scale. Consent is required and you may only clone voices you have the right to use. For an institution, that means an agreement with each instructor whose voice is cloned.

Credits count generated audio: about 1,000 credits per minute. A 30 minute module is about 30,000 credits.

Starter at ₹499 a month gives 100,000 credits, about 100 minutes, with commercial usage rights, which covers three modules a month. Creator at ₹999 gives 250,000 credits. Startup at ₹4,999 gives 1,500,000 credits, about 1,500 minutes or 25 hours, so about fifty 30 minute modules a month. Scale at ₹29,999 gives 10,000,000 credits, about 10,000 minutes, for a full catalogue in several languages. Annual billing gives 2 months free.Starter at $5 a month gives 100,000 credits, about 100 minutes, with commercial usage rights, which covers three modules a month. Startup at $49 gives 1,500,000 credits, about 1,500 minutes or 25 hours, so about fifty 30 minute modules a month. Scale at $299 gives 10,000,000 credits, about 10,000 minutes, for a full catalogue in several languages. Annual billing gives 2 months free.

The Free plan gives 10,000 credits, about 10 minutes, with every voice and no card. The web app at platform.chariot.in downloads MP3 or WAV; the API returns WAV. Either file goes into any authoring tool or video editor that accepts audio. For long scripts and video lessons, use Studio.