NC
← All articles
Talking calculator — a tool for elderly users and kids learning numbers
🧮
Accessibility29 Apr 2026 · 7 min read

Talking calculator — a tool for elderly users and kids learning numbers

Real use case: a calculator that speaks numbers and operations. We learn through an example our users actually requested.

This tutorial came from a real user request: “I need a calculator for my mom (74 years old). Her vision is poor. When she presses a button, I want it to say out loud what she pressed and what the result is.”

This problem reveals something important about working with AI: the quality of the result depends on the quality of the prompt. It's not the same to ask “make me a calculator” and “make me a calculator for my 74-year-old mother whose vision is poor.”

A prompt that WORKS

Build me a calculator that reads numbers and operations
out loud. The user is older with poor vision — buttons should
be large and high-contrast. Colors: dark background, white text.

When they press a number button — silent (display only).
When they press plus, minus, times, divide — read out loud
the entire number entered so far + the operation (e.g.
"three hundred fifty two times").
When they press =, read the result out loud ("equals
one thousand seven hundred sixty").

Speech must be in Serbian, female voice.
The display must show the full calculation — don't clear it.

What's specific here

  1. User context is in the prompt— “older, poor vision” → AI infers it needs large fonts and high contrast.
  2. Specific rules for when things happen— “silent for numbers, loud for operations”. No room for assumption.
  3. A concrete pronunciation example— “three hundred fifty two times”. Now the AI knows to convert digits to words (localization).

The technical thing you MUST know

Browsers (Chrome, Firefox, Edge) don't have a native Serbian voice in window.speechSynthesis. If you don't tell the AI anything specific, it will likely use a Russian voice as fallback — and your mom will hear something that sounds like a foreigner learning Serbian.

Solution (when you need a natural voice):tell the AI to use ElevenLabs Multilingual v2 model. It's cloud TTS with a native Serbian voice via the multi-language model. Free tier: 10,000 chars/month — enough for a calculator.

Tell the AI: "For speech use the ElevenLabs API instead of
window.speechSynthesis. API key is in env variable
ELEVENLABS_API_KEY (I'll add it). Voice ID:
21m00Tcm4TlvDq8ikWAM. Multilingual v2 model."

Our Skills system helps you

This specific case (TTS for Serbian) is already in our Skills database. When you mention “voice”, “reads”, “speaks”, or “TTS”, the AI automatically gets all this information without you having to write it manually. You don't need to know the technical details — the system knows for you.

What to change for a different user

  • For a child learning numbers — add “After each operation, briefly explain what happened”
  • For a deaf user — skip TTS, add large visual animations and strong haptic feedback (vibrate API)
  • For business use — add calculation history, CSV export, undo/redo