Disguise a voice in a recording VOICEGUARD

What a pitch shift hides, and what it leaves alone

Shifting the pitch moves the whole spectrum at once, including the fundamental frequency, the part of a voice a listener reads first and most confidently. Four semitones is usually enough that somebody who does not know you will not place the voice, and enough that a voice which read as clearly male or clearly female stops doing so.

It leaves everything else exactly where it was. Your accent, your vocabulary, the phrases you reach for, where you put your pauses, how you laugh, how you clear your throat. Somebody who knows you well often recognises a shifted voice within a sentence, because they are not recognising the pitch.

And the shift is reversible. It is one number. Anybody who suspects a recording has been pitched can shift it back by the same interval and hear the original, in this tool or any other. Treat this as a disguise from a listener who is not looking for you, not as anonymity from somebody who is.

Why the recording changes length too

The shift is done the way a tape machine does it: the samples are played faster or slower, so pitch and duration move together. Four semitones down makes a clip about a quarter longer; four up makes it about a fifth shorter. That is the honest consequence of the simplest method, and it works in your favour here, because the tempo and rhythm of somebody's speech is itself identifying and this moves it.

The robotic option multiplies the signal by a 55 Hz tone, which scatters every frequency in the voice into a pair either side of it. It sounds mechanical, it disguises further, and it costs some intelligibility. Listen to the result before you rely on it: a disguise that makes the words unclear has defeated the reason you were recording.

What the exported file carries

The download is a plain WAV written here from the audio samples, with a forty-four byte header and nothing else. It carries no device name, no recording date, no original filename, no location, because there is nowhere in that header to put them; the metadata the original file had does not survive the conversion. If you are working from a recording you are keeping as well as disguising, AVSCRUB strips the metadata from the original without touching the audio.

What the export does keep is the background. A station announcement, a particular clock, the hum of one specific room: none of that is pitched out of recognition and some of it places you precisely. HUSH bleeps or garbles named words, SNIP cuts a section out, and if the recording has to stand up later, SPLICE shows what an edit leaves behind so you know what somebody else would find.

If what you need is genuine anonymity rather than a disguise, no amount of processing gets you there from a recording of yourself. A listener who has a sample of your real voice can undo the shift and compare; a listener who does not still has your words and your manner. The reliable answer is to have somebody else read the words, or to have them read by a synthetic voice, so that there is nothing of you in the file to recover.