Why we built Apni Awaaz

Many people in India can think, read and hear, but can't take part in a phone call. Other countries have built help for this. India hasn't yet.

143,759new oral cancer cases in India in 2022, about 1 in 10 of all cancers

70–80%of these cancers are caused by tobacco and areca nut

16 secondsof an old TV clip was enough to rebuild one ALS patient's voice

The problem

Every year, many people in India lose the ability to speak. Mouth and throat cancer, mostly caused by tobacco, often ends in surgery that removes the voice box or tongue. ALS, Parkinson's, stroke and breathing tubes in intensive care take speech away too. These people can still hear and think, but they can't answer, not even on a call.

Many others are deaf or hard of hearing, and miss what's said on a call or at a shop counter.

Today's speech aids sound robotic, are slow to use and rarely speak Indian languages. Few apps caption calls in Indian languages. So many people simply stop talking to anyone outside their family.

What other countries do

In the US and the UK, a person who can't hear or speak can still make an ordinary phone call, through a relay service: an assistant or a text service sits between the two people and passes the conversation both ways.

United States

Since 2000, every phone company must connect the number 711 to a relay service. Speech-to-speech relay re-voices a caller whose speech is hard to understand, and internet relay users get ordinary ten-digit phone numbers.

United Kingdom

Relay UK, run by BT and overseen by Ofcom, lets people type or read calls. Hearing callers dial 18002 before a number to reach a relay user, and a relay assistant introduces the call.

India

In 2017 TRAI asked about relay-style services, and its 2018 recommendations set up a steering committee and help desks. India still has no national relay service.

What Indian law says

Rights of Persons with Disabilities Act, 2016. Section 42 says persons with disabilities must have access to information and communication technology.
Inclusive KYC and help desks (DoT, July 2025). Telecom companies must make sign-up and support accessible to persons with disabilities.
Digital Personal Data Protection Act, 2023, and Rules, 2025. A voice is personal data. Apni Awaaz keeps as little as it can and deletes everything when you ask (see our privacy policy).
IT Rules amendment on AI content (February 2026). AI-made audio must be labelled. That's why every Apni Awaaz call starts by saying the voice is made by AI.

The right to access exists on paper. Apni Awaaz is our attempt to make it work on an ordinary Android phone, in Indian languages.

The people who inspired us

Jennifer Wexton, a US Congresswoman, lost her voice to a brain disease and in 2024 gave a speech in Congress in an AI copy of her old voice.
Paul Francoeur, an ALS patient in Vermont, had his voice rebuilt from just 16 seconds of a 2012 TV interview, because his family had no other recordings.
Sapna Rana, a radio presenter in India for two decades, got her voice back after ALS took it, and calls for cheaper communication devices.

All of them used foreign tools built mainly for English. We want every Indian patient to have the same chance, in their own language.

What the research taught us

Long recording sessions put people off. In one voice banking service for motor neurone disease, under half of patients were told about voice banking, 1 in 4 finished it and 1 in 10 used it. Of 61 laryngectomy patients offered it in another study, 18 recorded enough and 7 used it soon after surgery.
A few seconds can now be enough. New voice cloning models can copy a voice from a short sample. Apni Awaaz needs as little as 5 seconds of speech, so it can be recorded at a hospital before surgery.
Your voice is part of who you are. Research on personalised voices found they help people keep their identity and stay in touch with others.

How we built it responsibly

A cloned voice unlocks only through a partner hospital, which records whose voice it is and who consented. We never recreate the voice of someone who has died.
Every call announces that the voice is made by AI.
Calls aren't recorded, and the other person's words are never stored.
Speech runs on Gnani.ai, built in India for Indian languages.