Accessibility · A LuxIA social impact project
Tocavox: a free app to speak with one tap, even in your own voice
Tocavox started with a concrete question from a family: how does someone who hears and understands everything, but can no longer speak, keep communicating — and could they do it in their own voice? We built a free web app where the person taps a phrase and the phone says it out loud, in a natural voice or in their own. It went from idea to production in under a week, and today it works in Spanish, English, French and Brazilian Portuguese at tocavox.com.
Visit tocavox.com →Problem
People who understand everything but cannot speak — after a laryngectomy or a stroke, with ALS or a tracheostomy — rely on a notepad or on someone guessing what they mean.
Result
A free web app in 4 languages: tap a phrase and the phone says it out loud, even in their own voice. From idea to production in under a week.
Can't speak? With Tocavox, your phone speaks for you: tap a phrase, it says it out loud and shows it in large letters. Free, no sign-up.
The context
Losing your speech while keeping your understanding is a very specific frustration: you know exactly what you want to say and have no way to say it. In the first days people write in a notepad; at a shop, in a taxi or at a doctor’s visit, that is slow and makes the other person read.
Existing tools are often paid, single-language or need special devices. We wanted something a person or their family could set up on their own, on the phone they already have, at no cost and without ads. And one more thing: it should be able to sound like the person, not like a generic voice.
How it works
Tocavox is a web app: it opens in the browser and can be added to the home screen. It works without signing up — ready-made phrases, a talking calculator and four languages; an account, created in one tap with Google, is only needed to personalize: your own phrases, typing any phrase and your own voice. The person sees big phrases organized by situation — home, health, taxi, buying, selling, greetings — and the ones they use most grow and move to the top. A calculator says amounts out loud, and “About me” builds phrases with their name, their area and who to call in an emergency.
Speech comes from an audio bank: each phrase is generated once, stored and reused. Base phrases are recorded and verified in advance, and the phone keeps them so they still play on a weak connection. For their own voice, the person records about 2 minutes in the app for a quick copy, or records around an hour with a guided recorder whose audio never leaves their device, to create a professional voice later.
Beyond the app, Tocavox publishes sourced guides for families and a YouTube channel with tutorials in all four languages: how to start in one minute and how to save your own voice before surgery.
Engineering decisions
Decision 01
The voice never receives digits
In testing we found that speech models didn’t just misread numbers: they made things up. A phone number turned into an emergency call nobody wrote. The fix was to spell numbers out before sending them — phone numbers digit by digit in groups, addresses and times in words — in all four languages. On screen the phrase still shows the digits.
Decision 02
Dual anti-fraud engine
So no one can use Tocavox to scam, every phrase a person writes goes through two engines. The first is a filter on our server with the signals of real scams: money, urgency, codes, threats or pretending to be a relative. If it is suspicious, the second is an AI judge with two models. A blocked phrase is not spoken with any voice. In tests with phrases used in scams, it blocked 93%.
Decision 03
Own voice with live consent, and reversible
To create a copy of their voice, the person reads an authorization with their name and the date, and a random sentence at the end: that proves it is them, live. If they record a new version, they compare it blind against the previous one on five phrases and decide; nothing is lost until they choose. People who already have a professional voice can connect their own ElevenLabs account; their key is stored encrypted.
Decision 04
Four languages without silent failures
Before adding a language, we made the code require content for every language in every place: if a translation is missing, the build fails instead of quietly showing Spanish. Translations and health guides were written by AI agents and validated by a separate, independent agent against the sources and the tone rules.
Decision 05
Health content that informs, never prescribes
Tocavox publishes guides for families (laryngectomy, stroke, ALS, tracheostomy, tongue surgery, voice rest). Every claim comes from an authoritative source that was opened and checked; none replaces the healthcare team or compares itself with other services. Tocavox is presented as one more support.
Decision 06
One tap, one phrase
Every new phrase first goes through the anti-fraud check, which takes a few seconds. In real-world testing, a person tapped “Save” several times while waiting, and the same phrase was saved 22 times, with one audio per copy. We fixed it the same day in two layers: the button locks and says “Saving…” until the server answers, and the server returns the existing phrase if a duplicate arrives, ignoring case and spacing. A nervous tap no longer turns into duplicates or into voices paid for twice.
What changes for the business
Tocavox is free and will stay free: no subscription, no paid version and no ads. It is a social impact project: LuxIA covers the servers and the voices, with help from people who choose to contribute, because someone who needs to speak should not have to pay to do it. Since October 2026 it has been helping people who cannot speak, without charging them a cent. For LuxIA it is also public proof of how we work: from idea to production in days, with verifiable quality, in several languages and with sensitive data handled carefully.
- From idea to production in under a week; Portuguese was added in full in a single day.
- 90 base phrases per language, in 4 languages, with audio generated and verified by transcription.
- Own voice from about 2 minutes of recording, or a professional voice from a long recording that stays on the person’s device.
- Six sourced guides for patients and families, in 4 languages.
- Dual anti-fraud engine: in tests with phrases used in scams, it blocked 93%.
- Video tutorials in 4 languages on its own YouTube channel.
Where else it applies
Accessibility
Augmentative communication on a phone, without special devices.
Customer service
Natural brand voices in several languages, reading numbers and amounts correctly.
Retail and counters
Calculators and quick replies spoken out loud.
Patient organizations
Free tools a family can set up on its own.
FAQ
Frequently asked questions about this case
Is Tocavox free?
Yes, and it will stay free. It is a LuxIA social impact project: no subscription, no paid version and no ads. LuxIA covers the servers and the voices, with help from voluntary contributions.
Can it speak in the person’s own voice?
Yes. About 2 minutes recorded in the app create a quick copy; a long recording can be used to create a professional voice on ElevenLabs, with a paid plan or for free if the person is approved for their Impact Program. What matters most is recording before the voice is lost. The voice only plays in the person’s account.
Can Tocavox be used to scam or impersonate someone?
We designed it not to. Every phrase a person writes is checked before it is saved and before it is said, with any voice; phrases that look like a scam are not spoken. Repeated attempts reach our team and, with an own voice, suspend it. No filter is perfect, but Tocavox is not a useful tool for that.
Does it replace therapy or medical care?
No. It is a support for everyday communication; health decisions are made with the healthcare team.
Can you build something like this for my organization?
Yes. The pattern — natural multilingual voice, own voice with consent, an audio bank and working on a weak connection — adapts to other contexts. Write to us and let’s talk.
Have a similar challenge?
Tell us your problem. We'll tell you how to solve it — and how long it takes.