Text to speech for accessibility: read anything aloud in 15+ languages

Someone who listens to their phone all day cannot use a voice that mangles their language. Chariot has 12 native Hindi voices and 2 native voices for each of Urdu, Bengali, Marathi, Telugu, Tamil, Gujarati, Kannada, Malayalam, Odia and Punjabi, plus English, German and Chinese: 15+ languages in all. Every voice can read every supported language.

In Chariot's own evaluation, naturalness (NISQA) scored 4.97 in Hindi, first among the systems tested, and the model had the lowest Hindi character error rate (7.73%) of the seven systems tested in Hindi. The clip below is Meera reading a Hindi weather report.

Real text is mixed. A Hindi news story has English names and terms in it, and a government notice has both. Chariot reads Hinglish and other mixes of an Indian language with English as one sentence, and reads Hindi written in Roman letters.

दिल्ली में आज शाम हल्की बारिश के साथ तापमान चौबीस डिग्री रहने का अनुमान है।

PlayPauseLoadingBufferingReplayRetry

0:05

Meera · Native Hindi

For a person who cannot check the screen, the voice is the only version of the document they have. If it reads a dose or an amount wrong, there is no second look. Chariot reads written text the way a person says it: numbers, dates, times, currency, units, abbreviations, phone numbers, account and order numbers, and URLs. "500 mg twice daily for 5 days" is read as instructions, "₹850" as an amount, an account number digit by digit.

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.

The clip below is Matt reading a prescription line: the dose, the frequency and the duration, each read as a person would say it.

Take five hundred milligrams twice daily for five days.

PlayPauseLoadingBufferingReplayRetry

0:04

Matt · American English

A reading feature that pauses for two seconds before every paragraph is one that people stop using. Chariot's streaming endpoint sends the first audio in about 75 ms (median, over WebSocket) and speaks sentence by sentence, so a long article starts within its first sentence while the rest is still being generated. Each request takes up to 1,000 characters, so send a paragraph at a time and the pauses disappear.

Speed is a request parameter from 0.5x to 1.5x. Give it to the listener rather than picking for them: people who listen all day tend to want it faster, and people new to listening want it slower.

The clip above is Susan, a British English voice, reading morning headlines, which is the kind of thing a news app sends through first thing.

No. Chariot is a text to speech API and a web app. Your app sends text and gets audio back. It does not read the operating system's interface, and it does not replace the screen reader built into a phone. What it does is give your app a voice for its own content: the article, the bill, the form, the notification, in a language the built-in voice may not handle well.

Most teams use it alongside the platform screen reader: the screen reader handles navigation, and the app's read-aloud feature handles the content. The decisions about how audio is announced, paused and interrupted stay in your app.

Text sent to Chariot and the audio it generates are processed in India. For an app that reads bank statements, prescriptions or government forms, that is the first thing a reviewer asks.

Chariot is built by one of four teams selected under the IndiaAI Mission to build foundation models. More than 10,000 people use it.

Credits count generated audio: about 1,000 credits per minute. The Free plan gives 10,000 credits, about 10 minutes, with every voice and no card, which is enough to wire up the streaming endpoint and test real documents.

Starter at ₹499 a month gives 100,000 credits, about 100 minutes, with 3 concurrent requests. Startup at ₹4,999 gives 1,500,000 credits, about 1,500 minutes, with 10 concurrent requests. Scale at ₹29,999 gives 10,000,000 credits, about 10,000 minutes, with 25 concurrent requests. Annual billing gives 2 months free.Starter at $5 a month gives 100,000 credits, about 100 minutes, with 3 concurrent requests. Startup at $49 gives 1,500,000 credits, about 1,500 minutes, with 10 concurrent requests. Scale at $299 gives 10,000,000 credits, about 10,000 minutes, with 25 concurrent requests. Annual billing gives 2 months free.

POST https://api.chariot.in/v1/tts returns a WAV file; the streaming endpoint sends raw 16-bit PCM at 44.1 kHz mono. There are Python examples. The web app at platform.chariot.in lets a non-developer test a document in any voice and download MP3 or WAV.