Tone generator.

Because pink has equal energy in every octave and white does not. An octave near the top of the spectrum is thousands of hertz wide and an octave near the bottom is tens, so a signal with equal energy per hertz — which is what white is — dumps almost all of its power into the top octave. Since both a spectrum analyser and human hearing divide the world into octaves, pink is the signal that reads flat on both, which is what makes a deviation from flat mean something about the room rather than about the noise.

Because it was measured, and the numbers are on this page. Averaging twenty transforms over twenty seconds of signal and summing the power in six consecutive octaves, the pink generator is flat to within 0.5 dB across the whole range, while the white generator rises by 14.8 dB over the same six octaves — which is the textbook +3 dB per octave and confirms the measurement as much as the generator. It matters: if a tuning tone were a decibel off, you would EQ a rig to compensate for the tone instead of for the room.

Quieter than you think, and turn the system down before you press play. Generated tones have no dynamics whatsoever — no crest factor, no transients, no gaps — so a level that feels normal for music is genuinely loud for continuous noise and can damage a tweeter at the top of a sweep. For tuning, a level where you can still hold a conversation is plenty; the analyser does not need it loud, and neither do your ears.

There is nothing to upload — the signal is generated in your browser from arithmetic rather than fetched from anywhere, and the download is written in the same tab. The page makes no network request once it has loaded.

Because that is what a digital mixing desk, a DSP processor and most club playback chains run at, so a 48 kHz file passes through without being resampled. Playback uses whatever your own audio hardware is running at, which is why the tool generates the signal fresh for each rather than resampling one version into the other.

A logarithmic sweep rises by a fixed number of octaves per second, so it spends as long travelling from 40 to 80 Hz as it does from 5 to 10 kHz. A linear sweep rises by a fixed number of hertz per second, which means it covers the entire bass range in the first fraction of a second and then crawls through the top. Since almost every room problem is in the bass, a linear sweep skips the interesting part.

Playback here already loops, and the downloaded noise files loop seamlessly because noise has no phase to match up. A sine will loop cleanly only if its length happens to be a whole number of cycles, and a sweep is not meant to loop at all — the jump from 20 kHz back to 20 Hz is a click.

To stop the tone being a click. A waveform that begins at full amplitude on the first sample is a step function, and a speaker asked to reproduce a step does something violent and broadband before it settles into the tone. Twenty milliseconds of cosine at each end removes it entirely and is far too short to affect any measurement.

Yes. They are generated signals with nothing to license, so use them for sound checks, installs, measurements, teaching, whatever you like.