Frequency detector.

No. The stream is read frame by frame to measure it and nothing is stored anywhere — not in the page, not on disk, and certainly not on a server. The microphone is only opened when you click the button and it is released the moment you stop, so nothing is held while the page sits idle. The page makes no network request at all once it has loaded.

If it reads 50 Hz you are on a 50 Hz mains supply and something in the chain has a ground loop or an unbalanced connection; 60 Hz is the same fault in North America. If it reads 100, 150, 200 or 300 Hz you are hearing the harmonics, which is what makes it a buzz rather than a hum and usually means a dimmer, a cheap power supply or a laptop charger sharing the circuit. The tool names all of these rather than leaving you to recognise the number.

Accurate to a fraction of a hertz on a steady tone. The transform on its own gives bins about 2.9 Hz apart, which is not enough to tell 50 Hz from 60 Hz reliably once real-world spread is involved, so the peak is refined by fitting a parabola through it and its neighbours. On a moving or noisy signal the reading moves with it — that is honest rather than a fault, and the hold button is there for the moment you get a clean look.

Because most sounds are not one frequency. Music, speech and room noise all contain many at once and the loudest changes moment to moment, so the headline number follows whatever is currently winning. On a steady hum or a sustained note it settles immediately. The table underneath shows the four strongest things present at once, which is usually more informative than the single winner.

Yes — drop the file and the strongest frequency across the whole thing is named as a note with its error in cents. It works well on a bass note, a stab, a held vocal, or anything with a clear pitch. It is not a key detector: a full track has a whole arrangement in it and the loudest single frequency is usually the kick. For a whole track, the key and BPM finder is the right tool.

Because a fixed number of hertz means completely different things at different pitches. Three hertz at 50 Hz is most of a semitone; three hertz at 5 kHz is imperceptible. Cents are hundredths of a semitone, so a hundred cents is one semitone wherever you are in the spectrum, and about five cents is where a trained ear starts to notice.

Yes, and it is arguably the better device for it, since the point is to carry it to the thing that is humming. The browser will ask for microphone permission the first time. A phone microphone rolls off in the deep bass, so a very low rumble may read quieter than it is — the frequency will still be right.

Because both are designed to remove steady, predictable sounds, which is exactly what you are pointing this at. Left on, they would quietly delete the hum you are trying to identify, and automatic gain control would move the level under the measurement. The stream is opened with all three disabled so the microphone hands over what it actually hears.

For mains hum, fix the cabling rather than the EQ — a notch filter hides a ground loop but the noise is still being injected and it will come back louder on a bigger system. For a room mode, move the speakers or the listening position before reaching for EQ, because a mode changes with position and EQ does not. For feedback, notch narrowly and only as much as you need. The tone generator is useful for confirming a fix by ear afterwards.