Speech Synthesis Fingerprinting

Speech synthesis fingerprinting reads the list of text-to-speech voices available through the speechSynthesis API. Voice sets are OS-dependent and vary significantly across platforms and locales.

How it works

Neoprint calls speechSynthesis.getVoices() and records each voice's name, language, whether it's a local voice, and whether it's the default.

const voices = speechSynthesis.getVoices()
const data = voices.map(v => ({
  name: v.name,       // "Zosia"
  lang: v.lang,       // "pl-PL"
  localService: v.localService,  // true = OS voice, false = cloud
  default: v.default,
}))

Why it's effective

OS	Typical voice count	Notable voices
macOS	60-80 local	Samantha, Alex, Zosia (Polish)
Windows	5-20 local	David, Zira, Mark
iOS	60-80 local	Same as macOS
Android	5-30	Google TTS voices
Linux	0-5	espeak voices (if installed)

Users who install additional language packs get more voices, further differentiating their profile.

Entropy and stability

Property	Value
Entropy	~10 bits
Stability	0.90
Typical duration	50-60ms

The high duration is due to async voice loading. Neoprint waits up to 300ms for the voiceschanged event if voices aren't immediately available.

Cross-browser considerations

Chrome exposes more voices as localService: true than Safari on the same OS (180 vs 68 in testing). Speech is excluded from crossBrowserId and stableId because voice lists differ too much between browser engines on the same OS, and Safari returns 0 voices in private browsing.

Role in incognito detection

Safari private browsing returns 0 voices from speechSynthesis.getVoices(), while normal mode returns 68. This is one of the strongest signals used by neoprint.detectIncognito() for Safari.

Usage

const fp = await neoprint.get({ collectors: ['speech'] })
const voices = fp.components.speech.value

console.log(voices.length)          // number of available voices
console.log(voices[0].name)         // "Zosia"
console.log(voices[0].lang)         // "pl-PL"
console.log(voices[0].localService) // true