Numbers for Text-to-Speech (TTS)

TTS engines speak natural language — feeding them "12345" sounds robotic. Feed them "Twelve Thousand Three Hundred Forty Five" instead.

The Problem

Raw digits make speech engines guess. Some engines spell each digit, some pause awkwardly, and some fail entirely on locale-specific currency or number grouping.

Before and After

Raw digits:  12345
Spoken text: Twelve Thousand Three Hundred Forty Five

Browser: Web Speech API

import { toWords } from 'to-words';

const utterance = new SpeechSynthesisUtterance(toWords(12345, { localeCode: 'en-US' }));
speechSynthesis.speak(utterance);

Multilingual TTS

import { toWords } from 'to-words';

function speakAmount(amount, localeCode, lang) {
  const utterance = new SpeechSynthesisUtterance(toWords(amount, { localeCode }));
  utterance.lang = lang;
  speechSynthesis.speak(utterance);
}

speakAmount(1500000, 'hi-IN', 'hi-IN');
speakAmount(12345, 'ar-AE', 'ar-AE');

Node.js: Azure Cognitive Services

import { toWords } from 'to-words';
import sdk from 'microsoft-cognitiveservices-speech-sdk';

const text = toWords(1500000, { localeCode: 'hi-IN' });
// "पंद्रह लाख"

const synthesizer = new sdk.SpeechSynthesizer(speechConfig, audioConfig);
synthesizer.speakTextAsync(text);

Why it matters

Without spoken-number conversion, a TTS engine will either spell out digits or guess a pronunciation. to-words outputs clean, grammatically correct words that every TTS engine pronounces correctly.

Related

• Getting Started
• Hindi locale
• Arabic locale
• Framework integration