Audionix — the engine behind the audio tools

Audionix is an audio engine — the piece of software that does the arithmetic when a recording is pitched, stretched or retuned. It is not a plugin you install and not a service you upload to. It is compiled to WebAssembly and shipped inside the web page, so the processing happens on the machine in front of you.

Every audio tool on AiZoneHub is a face on this one engine. The pitch changer, the tempo changer, the equalizer, the cutter and the rest are different controls over the same code, which is why a track loaded into one of them behaves identically in all of them.

Audionix is made by AiZoneHub and is free to use, with no account, no upload and no limit on how many files you put through it.

A voice carries two separate things at once. There is the pitch — the note being sung — and there is the timbre, the fixed set of resonances of a particular throat, mouth and chest that make a voice recognisable as one person and not another.

Most pitch shifting drags both up together. Move a song up three semitones so a singer can reach it and the singer turns into a child; move it down and they turn into a giant. The notes are right and the person is gone. That artefact is the reason so many free pitch tools are unusable for anything but a joke.

Audionix treats the two as two controls. Pitch can move while timbre stays where it was, which is what makes a key change usable on a vocal — or timbre can move on its own, with the notes untouched, which is a different effect entirely and one most tools cannot do at all.

A pitch shift is done in two moves: stretch the recording in time by some ratio, then resample it by the inverse of that ratio. The stretch is where all the difficulty lives, and Audionix does it with a phase vocoder — the audio is cut into overlapping windows, each is taken into the frequency domain, the phases are advanced to the new timing, and the windows are added back together.

The catch is that one window size cannot serve the whole spectrum. A window long enough to resolve a bass note smears a consonant; a window short enough to keep transients sharp cannot tell two low notes apart. Audionix therefore runs two bands at two resolutions — a long window for the low end, a short one above it — and splits them in the frequency domain, bin by bin, rather than with a filter. A filter would let the same partial travel down both paths and partially cancel where they meet; measured, that cost a notch several hundred hertz wide whose position moved with the pitch setting. Splitting at bin granularity, with a power-complementary taper across the seam, removes it.

Timbre is a separate step applied afterwards: the spectral envelope — the shape the resonances draw across the spectrum — is warped on its own, independently of where the notes went. That separation is the whole design, and everything else follows from it.

The engine is written in C++ and compiled to WebAssembly, with a SIMD build for browsers that support it and a plain build for those that do not. The browser picks one at load time.

Running on the device is a privacy decision before it is a technical one. An unreleased demo, a client’s recording or a private voice note stays on your own machine, because there is no upload step in which it could go anywhere else. It also means the tools keep working with the network off, and that there is no queue and no file-size limit imposed by somebody’s server budget.

The browser build is held to the native one as its reference: the same input through both produces output that agrees to within a thousandth of a decibel. The engine’s real-time path and its export path are checked against each other the same way, across eighteen block sizes and nine pitch and timbre pairs, and are required to match exactly rather than closely.

Audionix is spelled with an x. It is the audio engine built by AiZoneHub, a web project, and it has nothing to do with Audionic, spelled with a c, which is a consumer electronics company that makes speakers and headphones. Different companies, different products, one letter apart.

If you arrived here looking for speakers, you want the other one. If you arrived here looking for a way to change the key of a song without wrecking the vocal, you are in the right place.