Online Spectrum Analyzer

A fast Fourier transform takes a short slice of sound and splits it into equally spaced frequencies called bins. The spacing between them, the bin width, is the sample rate divided by the FFT size: at 48 kHz, 8,192 points gives bins 5.86 Hz apart, and 32,768 points gives 1.46 Hz. A tone that falls between two bin centres is shared between them, which is why this page fits a curve through neighbouring bins to place a peak more precisely than the spacing alone allows.

Finer frequency detail needs a longer slice of sound. 32,768 points at 48 kHz is 683 milliseconds of audio, so a note has to last that long to be fully represented, and a quick change is smeared across the window. 2,048 points is 43 milliseconds and reacts almost at once, but its 23.4 Hz bins cannot separate 50 Hz from 60 Hz. Use a large size for steady sounds such as hum or a held note, and a small one for speech, drums, or anything that moves. Broadband noise also reads lower per bin as the bins get narrower, while a pure tone stays at the same level.

No. dBFS is decibels relative to digital full scale, where 0 dBFS is the largest value the converter can hold, and every reading here is on that scale, with a full-scale sine wave at 0 dBFS. dB SPL is sound pressure in the room, and getting there needs the microphone's sensitivity, the preamp gain, and the operating system's input volume, none of which a browser can know. Turn the input volume down and every reading drops while the room stays exactly as loud. The shape of the spectrum and the relative heights of its peaks are meaningful; the absolute numbers are not loudness.

Browsers clean up microphone audio for calls before any page sees it. Noise suppression removes steady sounds such as fans and mains hum, which is often exactly what you are trying to find. Echo cancellation subtracts whatever the speakers play, which can notch out a test tone. Automatic gain control raises and lowers the level as you watch. All three start off here, and the line under each switch shows what the browser says it actually applied, since some browsers ignore the request on some devices.

Set the FFT size to 16K or 32K so the bins are under 3 Hz apart, and use the spectrum view on the logarithmic axis. Mains hum sits at 50 Hz in Europe, Africa, most of Asia, and Australia, and at 60 Hz in the Americas and parts of Asia, with harmonics at whole multiples: 100, 150, and 200 Hz, or 120, 180, and 240 Hz. A strong 100 or 120 Hz line over a weak fundamental usually points at a transformer or a power supply rather than a ground loop. Watch the peak list as you unplug or move cables one at a time.

Built-in microphones are tiny, and they are filtered on purpose to keep out handling noise, wind, and the device's own vibration, so most of them fall away steeply below about 100 Hz. The analyser shows what the microphone delivers, not what is in the room. To see real bass, use a measurement microphone or a decent USB or interface microphone, and keep voice processing off, since some of it adds its own low cut.

Time runs from right to left, frequency from bottom to top, and brightness is level, from the page's ink for silence through lime to near white for the loudest. A steady tone draws a horizontal line, a voice draws a stack of parallel curves (its harmonics), a whistle glide draws a slope, and a clap is a short vertical stripe. It keeps filling while you look at another view, so switching to it shows the last few dozen seconds.

For a steady tone, very. Each peak is fitted with a parabola through the three bins around it, and tested on synthetic tones through the same Blackman window the browser uses, the estimate landed within 0.007 of a bin, which is under 0.04 Hz at the default 8,192 points and 48 kHz. Real sound adds noise, vibrato, and nearby partials, so treat it as accurate to a small fraction of the bin width shown. The audio clock matters too: a converter running 50 parts per million fast reads every frequency 50 parts per million high. And the loudest peak is often a harmonic rather than the note being played.

Yes, as a comparison tool. For feedback, raise the gain slowly until the system starts to ring, freeze the display, and read the loudest peak: that is the frequency to cut on the equaliser, with its nearest note alongside. For a room mode, play pink noise or a slow sine sweep and look for peaks that stay put as you move the microphone, usually below 300 Hz. Because the microphone is uncalibrated, compare positions and settings against each other rather than reading the numbers as absolute.

No. The analyser reads the microphone or the file inside this tab with the Web Audio API and keeps only the latest frame for drawing. Nothing is sent to a server, nothing is saved in the browser, and nothing is recorded. A PNG or CSV exists only when you press export, and it goes straight to your device. Stopping the microphone releases it, and the browser's recording indicator should go out. A local file plays from your own device and is never uploaded.