Audio fingerprint (OfflineAudioContext)
Renders a sound offline and reads tiny numeric differences.
In one line
Can make a silent test sound to tell your device apart from others
Renders a sound offline and reads tiny numeric differences.
Audio stack differences give a device specific value without ever playing sound or asking permission. LinkedIn uses OfflineAudioContext(1,44100,44100).
What it can and can't do
What the user does: Nothing. It happens when the page loads
It can
- Process a short sound in memory without playing it
- Measure tiny number differences that depend on your device and browser
- Combine that number with other signals to recognise you later
It cannot
- Play sound through your speakers
- Record from your microphone
- Ask you first, since there is no pop up
- Work as an ID on its own, because many similar devices give the same result
Browser support
| Browser | Status | First version | Released | Note |
|---|---|---|---|---|
| Chrome | Supported | 35 | 2014-05-20 | None |
| Edge | Supported | 12 | 2015-07-29 | None |
| Firefox | Supported | 25 | 2013-10-29 | None |
| Safari | Supported | 14.1 | 2021-04-26 | None |
| Safari on iPhone | Supported | 14.5 | 2021-04-26 | None |
| Chrome on Android | Supported | 35 | 2014-05-20 | None |
| Samsung Internet | Supported | 3.0 | 2015-04-10 | None |
How a site asks for it
| API | new OfflineAudioContext(1, 44100, 44100) |
| Browser compat keys | api.OfflineAudioContext |
| How access is granted | none |
| What the user sees | None (no browser prompt) |
| Needs a secure context (HTTPS) | No |
| Needs a user gesture | No |
| How long it lasts | Not listed |
| How to take it back | Not listed |
| Inside an iframe | Not listed |
Typical flow
- Script reads the value
- Value joins the fingerprint payload
Data you get
What comes back: Audio fingerprint (OfflineAudioContext)
Shape: hash
How often: static per setup
| Field | Type | Unit | Range | Precision |
|---|---|---|---|---|
audio_sum | float | None | hash of samples | float precision |
Technical limits
- Noise added by some privacy browsers
The story
What it is
A page plays a sound silently in memory and reads tiny number differences that vary by device.
For you
You hear nothing and see no prompt. The page just does some math that a sound chip would do, and the answer helps it recognize your device later.
For makers
Trackers and fraud teams add this to a bundle of signals to build a device ID. It is cheap, needs no permission, and is hard for people to notice. The real feature exists so audio apps can process sound offline.
Where you see it
| Type | Example | Apps |
|---|---|---|
| music | A web music studio renders a finished track to a file offline, much faster than playing it back. | BandLab, Soundtrap |
| games | A browser game builds echo or reverb effects ahead of time so sound effects play instantly in battle. | None named |
| security | A bank's fraud script plays a silent tone and compares the result with the last visit to confirm it is the same device. | None named |
| ads | An ad tech script adds the audio result to a device ID to recognize you across sites without cookies. | None named |
| analytics | A visitor counting tool uses the audio hash to avoid double counting the same person who cleared cookies. | FingerprintJS |
| creative | A podcast editor in the browser exports a mixed episode by rendering all tracks offline. | None named |
| accessibility | A page prepares sound effects for screen reader users before they hear them. | None named |
Risks
- Works without asking
- Hard to reset
- Browsers are adding noise to fight it
What the platform says
- MDN: OfflineAudioContext. Renders audio faster than real time into a buffer, without playing it to the speakers, for effects and analysis.
Good uses and abuse
Legitimate uses
- Audio rendering tests
Known abuse
- Fingerprint input
Scores
| Score | Value | Meaning |
|---|---|---|
| Sensitivity | 2 of 5 | How revealing the data is |
| Utility | 3 of 5 | How useful it is to honest apps |
| Friction | 0 of 5 | How much the user has to do |
| Fingerprint bits | 5 | Rough identifying entropy, when known |
A real example: LinkedIn
Collector key(s): audio. BleepingComputer confirms audio.
Related capabilities
Related
- APFC / DNA device fingerprint (abuse prevention features). Can collect about 48 device traits and send them in secret with later requests
Sources
- MDN OfflineAudioContext
- BleepingComputer: LinkedIn scans for 6,000+ Chrome extensions
- mdp/linkedin-extension-fingerprinting (LinkedIn script snapshot and ID list)
How well checked: Based on official documentation.