Audio Spectrum Analyzer Online

See the frequencies in any sound around you as they happen. This real-time audio spectrum analyzer listens through the microphone in your phone, tablet or laptop and draws what it hears: a fine FFT spectrum, one-third-octave or octave bands the way a real-time analyzer (RTA) shows them, or a scrolling spectrogram. As a frequency analyzer it names the loudest frequency, its level and the nearest note, and it can hold peaks, average, freeze the picture and save it as a PNG or the figures as a CSV. Nothing is recorded or sent.

Before you start: the chart opens on an example made in your browser, a sung A3 with its harmonics over pink noise. Press Start and allow the microphone to replace it with what your microphone hears. Levels are in dBFS, measured down from the loudest level the input can carry.

Live spectrum

Showing an example. Press Start and allow the microphone when your browser asks.

View
Loudest frequency
–
Nearest note
–
Its level
–
Input peak
–

Point at or tap the chart for the frequency and level there.

Measured in your browser. Nothing is recorded, stored or sent, and Turn microphone off releases the microphone at once.

How to read a spectrum analyzer

Three views of one sound

On the spectrum, frequency runs left to right and level bottom to top, and every point is one step of the FFT. A steady tone draws a narrow spike, and a sung or played note draws a row of spikes at whole-number multiples of its pitch, its harmonics. The top of a spike is the tone's level: a full-scale sine reads 0 dBFS.

The one-third-octave and octave views add up every step inside each band, the way a real-time analyzer (RTA) shows sound. Each band covers the same share of an octave, so pink noise, which carries the same energy in every octave, draws a flat row of bars. On the spectrum line the same pink noise falls 3 dB with every octave, because each step is the same width in hertz while each octave up holds twice as many of them.

The spectrogram scrolls from right to left with the newest moment at the right edge, frequency bottom to top and level as color. To study a recording instead of a live sound, the audio spectrogram draws a whole file.

Using it as a room analyzer

Play pink noise from the signal generator through your speakers at a comfortable level, hold the microphone at your listening seat, pick the one-third-octave view and set averaging to Long. A band that stands well above its neighbors in the bass is often a room mode, which the room mode calculator predicts from the room's size. Move the microphone a little and read again: room bass changes from seat to seat.

A phone or laptop microphone is made for speech, its response is not flat, and the browser does not report it, so the microphone's own coloring is in every reading. Compare readings from one microphone with each other. For a room measurement to rely on, use a measurement microphone with its calibration file in measurement software. To read how loud the room is in dB SPL, the decibel meter reads the same microphone and can be calibrated against a real meter.

The loudest frequency

The readout finds the loudest step between 20 Hz and 20 kHz and fits a curve through it and its two neighbors, so a steady tone's frequency is read to a small fraction of a step and its level to about a tenth of a decibel. The nearest note is equal-tempered with A4 at 440 Hz. To tune an instrument, the online tuner reads pitch with a method made for it, which finds the fundamental even when a harmonic is louder.

Window size: detail against speed

Each spectrum is measured from the last window of samples. A longer window splits the spectrum into finer steps and separates the bass bands, and a shorter one follows a quick change sooner. A step is the sample rate divided by the window size. A one-third-octave band narrower than six steps, the width of the window's main lobe, shares a tone with its neighbors, so the chart draws it fainter; the last column is the lowest band each window separates at 48 kHz.

WindowStep at 44.1 kHzStep at 48 kHzLength at 48 kHzLowest band separated
2,048 points21.53 Hz23.44 Hz43 ms630 Hz
4,096 points10.77 Hz11.72 Hz85 ms315 Hz
8,192 points5.38 Hz5.86 Hz171 ms160 Hz
16,384 points2.69 Hz2.93 Hz341 ms80 Hz
32,768 points1.35 Hz1.46 Hz683 ms40 Hz

What a spectrum analyzer measures

Any sound can be split into frequencies, each with its own level, and a fast Fourier transform (FFT) does the splitting from a short stretch of samples. A spectrum analyzer runs one FFT after another on the newest samples, so the picture follows the sound as it changes.

This page asks the browser's own analyzer for each spectrum: it takes the last window of samples from the microphone, fades its edges with a Blackman window and measures it, many times a second. The page does the rest itself. It puts a full-scale sine at 0 dBFS, averages in power, holds peaks, applies the weighting curve and adds up the bands.

The window length decides what the picture can resolve. A window of 8,192 samples at 48 kHz lasts 171 ms and splits the spectrum into steps of 5.9 Hz. A longer window separates the low bands but reacts more slowly; a shorter one reacts sooner with coarser steps.

The ideas behind the controls

FFT
The fast Fourier transform: the arithmetic that turns a stretch of samples into the level at each frequency.
Step (bin)
One point on the frequency axis: the sample rate divided by the window size. At 48 kHz and 8,192 points a step is 5.9 Hz wide.
RTA
A real-time analyzer: the spectrum added up into one-third-octave or octave bands, so pink noise reads flat and a room's bass peaks stand out.
dBFS
Decibels below full scale, the loudest level the input can carry. A full-scale sine reads 0 dBFS, so every level on the chart is 0 or below.
Averaging
Spectra averaged over time, in power, so noise settles into a steady line. Short follows a change within a fraction of a second, Long over a couple of seconds.
Peak hold
The loudest level each frequency or band has reached since the last reset, drawn as a thin line above the live one.

Spectrum analyzer FAQ

  1. What is an audio spectrum analyzer?

    An audio spectrum analyzer shows how much of a sound sits at each frequency: bass on the left, treble on the right and level up the side. This one measures the sound from your microphone many times a second with a fast Fourier transform (FFT), so the picture moves with the sound, and it shows the same measurement as one-third-octave or octave bands or as a scrolling spectrogram.

  2. How do I see the frequencies in a sound?

    Press Start spectrum analyzer and allow the microphone, then make or play the sound near it. The tallest spike on the spectrum is the loudest frequency, and the tiles above the chart name it, its level and its nearest note. Freeze holds the picture, so you can point at or tap any part of it and read the frequency and level there.

  3. How do I use it as an RTA (real-time analyzer)?

    Play pink noise through your speakers (the signal generator on this site makes it), put the microphone where you listen, pick the 1/3 octave view and set averaging to Long. Pink noise carries the same energy in every octave, so a system that plays it evenly draws flat bars, and the bands that stand above or below the rest show where the speakers and the room add or take away. Read it as a guide: a phone or laptop microphone has a response of its own, and it is in every reading.

  4. Why does pink noise slope down on the spectrum but look flat in the bands?

    Pink noise holds half as much energy per hertz with every doubling of frequency, 3 dB less, and each octave up spans twice as many hertz, so every octave ends up with the same energy. The spectrum view draws one point per FFT step, every step the same width in hertz, so it shows the 3 dB fall per octave. The band views add up whole bands, so they show each octave's equal share as a flat row.

  5. What is the difference between a spectrum analyzer and a spectrogram?

    A spectrum analyzer shows the frequencies of one moment, or an average of the last few: one line or one row of bars. A spectrogram lays those moments side by side, so you see how the frequencies change over time. This page shows both for live sound from the microphone, and the audio spectrogram on this site draws a whole audio file.

  6. How accurate is the frequency reading?

    For a steady tone, close: the readout fits a curve through the loudest FFT step and its two neighbors, which places a clean tone within a hundredth of a step and its level within about a tenth of a decibel in the page's tests. At the default window of 8,192 points a step is 5.9 Hz at 48 kHz, so that is well under a tenth of a hertz. Noise around a tone makes the reading wander; averaging steadies it.

  7. Why are the levels in dBFS and not in dB SPL?

    A browser hands the page samples from the microphone, and how many dB SPL a sample stands for depends on the microphone, its gain and the operating system, none of which the page can read. dBFS is measured from the loudest level the input can carry: a full-scale sine reads 0 dBFS and everything quieter reads below it. For a sound level in dB SPL, the decibel meter on this site reads the same microphone and can be calibrated against a real meter.

  8. Should I use A-weighting?

    Leave it on Z, no weighting, to see the sound as the microphone measures it, which is what you want for tones, hum and room bass. A-weighting turns the deep bass and the top treble down roughly the way hearing does at moderate levels, about 39 dB down at 31.5 Hz, and C-weighting trims only the extremes. Both are the IEC 61672-1 curves the decibel meter uses.

  9. Why does the chart show sound when the room is quiet?

    Every microphone and input adds a little noise of its own, and a quiet room still holds ventilation, traffic and the hum of electronics, so the line never falls to nothing. The quietest part of the line is that noise floor. The mic test on this site measures it and says whether it is low enough for calls or recording.

  10. Does it work on a phone?

    Yes. Open the page in Chrome, Safari or Firefox, press Start and allow the microphone. A phone microphone is built for voices, so read the deep bass and the top treble with care, and hold the phone still: handling noise shows up as bursts in the low bass.

  11. Is the sound recorded or uploaded?

    No. The page measures the sound in your browser as it arrives and keeps only the numbers on the chart. Nothing is recorded, stored or sent, and Turn microphone off releases the microphone at once.