Free, in your browser
Spectrogram Online
See any sound as a spectrogram and read it precisely. Drop in a song, a voice memo, a field recording or the sound of a video, even one hours long, and this spectrogram draws the whole file; point at any line and it snaps to the peak, reading the measured frequency, the note and the level in dBFS. Or start the microphone and watch a live spectrogram of your voice. Everything stays on your device.
- no sign-up, nothing uploaded
- Free
- worst readout error on 14,400 test tones
- 0.004 Hz
- drawn whole in about 13 s in desktop Chrome
- 3 hours
- three frequency scales
- Log, linear, Mel
- FFT points, time or frequency detail
- 256 to 32,768
- PNG with axes and dBFS colour bar
- 1920 × 1080
What is a spectrogram?
A spectrogram is a picture of sound over time: time runs from left to right, frequency from bottom to top, and the colour of each point shows how loud that frequency is at that moment, here in dBFS. To make a spectrogram online, open an audio or video file in this spectrogram, or start the microphone, then point at any line to read its frequency, note and level.
Make an audio spectrogram in three steps
Audio to spectrogram in your browser, on a computer or a phone, from any file it can read or from the microphone.
Open a file or start the microphone
Drop in audio or a video of any length and its whole spectrogram is drawn. Or switch to the Microphone tab and start it.
Zoom in and point at a line
Wheel or pinch to zoom, drag to move, double click for the whole file, click to play from there. The spectrogram readout snaps to the nearest peak.
Save the spectrogram as PNG
1920 × 1080, with frequency and time axes, the dBFS colour bar, the file name and the settings you used.
A spectrogram that measures what it shows
An audio spectrogram is as useful as what you can read from it. Here each readout is measured from the sound itself, and every number on this page comes from tests on signals whose answers are known.

Read the measured frequency, to 0.004 Hz
Pointing at a line usually reads the row under the cursor, a bin away from the truth. With Snap to the peak, the spectrogram jumps to the nearest peak and measures its frequency from the phase of two frames, and its level in AES17 dBFS. On 14,400 test tones the worst error was 0.004 Hz and 0.003 dB.
Files hours long, drawn whole
Long recordings are usually cut short or played through in real time. Here several background workers read the file a chunk at a time, so memory stays small, and draw the full spectrogram; zoom in and that stretch is read again at one frame per pixel. A 60-minute MP3 is drawn whole in about 5 seconds in desktop Chrome, a 3-hour one in about 13.
Log, linear or Mel spectrogram
Music and voices live low, and a linear scale squeezes them into the bottom strip. Log gives every octave the same height, so the site’s 20 Hz to 20 kHz sweep is a straight line; linear spreads the top for hidden pictures and cutoffs; Mel, Slaney’s version, is the common default in speech and machine learning.
Choose time detail or frequency detail
A spectrogram trades one for the other: short frames place a drum hit to the millisecond, long frames split notes a few hertz apart. One slider moves between them, 256 to 32,768 points, and shows the frame length and the spacing it separates: at 48 kHz, 256 points is a 5.3 ms frame and 32,768 points is 683 ms.
See where the content stops: 16.7 kHz at 128 kbps
Lossy encoders drop the highest frequencies, and a clean edge is easy to miss by ear. The spectrogram finds the cutoff of the whole file and says whether it is a sharp edge, as lossy encoders leave it, or reaches the top. Our own MP3 encoder stops at 15.3 kHz at 96 kbps and 20.2 kHz at 320 kbps.
An audio spectrum analyzer for the whole file
Beside the spectrogram, the average spectrum of the whole file shows which frequencies carry the most energy and names the loudest one. Mains hum stands out as a line at 50 or 60 Hz with its harmonics stacked above at even steps. On white noise, the average spectrum reads within 0.025 dB of the true density.
Spectrogram samples: where to look
Twelve sounds from this site, each drawn and read by the same code as the spectrogram studio, with a note on where to look and what the readout said. Open any card in the studio to zoom in and read it yourself.
How to read a spectrogram: a practical guide
Spectrogram, spectrum and waveform
A waveform shows how the level of a sound moves over time. A spectrum shows which frequencies a stretch of sound holds, added up over that stretch. A spectrogram joins the two: it cuts the sound into short overlapping frames, takes the spectrum of each one (the short-time Fourier transform) and lays them side by side, so you see which frequencies sound when.
Here the average spectrum of the whole file sits beside the picture, so the audio spectrum analyzer view and the spectrogram view are one click apart.
How to read a spectrogram
Every shape on a spectrogram has a meaning once you know the patterns:
- 1A horizontal line is a steady tone: a whistle, a held note, a hum.
- 2A stack of evenly spaced lines is one note and its harmonics: a voice, a violin, a trumpet. The lowest line is the pitch you hear.
- 3A vertical line is a click or a hit: a drum, a consonant, a metronome tick.
- 4A haze across many frequencies is noise: wind, hiss, a fan, a cymbal.
- 5A flat top edge across the whole file is lossy encoding, or the bandwidth of the device that recorded it.
- 6Brighter is louder. The colour bar reads in dBFS, where 0 is full scale and −60 is 60 dB below it.
Point near a line and let Snap to the peak find it: the readout gives the measured frequency, the note with its cents and the level of the line itself.
Choosing spectrogram settings
The presets (Any sound, Speech in wideband or narrowband, Birdsong, Hidden pictures and Check the cutoff) set the whole spectrogram at once. To set it yourself:
- 1Time or frequency detail: 256 points for drums and clicks, 4,096 for most music, up to 32,768 to split close harmonics.
- 2Window: Hann for most sounds; Blackman-Harris keeps a loud tone from spreading over quiet ones; Gaussian for speech.
- 3Overlap: 50, 75 or 87.5%; more overlap draws a smoother spectrogram along time.
- 4Scale: log for music and voices, linear for hidden pictures, cutoffs and high birdsong, Mel for speech and machine learning.
- 5Brightness: automatic fits the colours to the file; set the top and bottom in dBFS to compare two files on the same colours.
Is my FLAC really lossless? Check the cutoff
A song converted from MP3 or AAC to FLAC or WAV keeps the lossy cutoff: the spectrogram shows a sharp top edge, level from start to finish. A full-band recording saved lossless usually reaches the top. The cutoff of the whole file is listed in the file facts beside the sample rate, bitrate and loudest frequency, and the Check the cutoff preset sets the spectrogram up to see it.
Where our own MP3 encoder (the audio compressor’s, constant bitrate) stops a full-band sound:
- 196 kbps: 15.3 kHz
- 2128 kbps: 16.7 kHz
- 3160 kbps: 17.4 kHz
- 4192 kbps: 18.8 kHz
- 5256 kbps: 19.5 kHz
- 6320 kbps: 20.2 kHz
A sharp edge looks like a lossy source, and the spectrogram says just that. A high-bitrate MP3 or AAC can also reach 20 kHz, so read a full top edge as a good sign rather than proof.
Birdsong and whistles on a spectrogram
Bird calls are short and high, so on a spectrogram they show as small marks, slides and trills well above a voice. Pick the Birdsong preset, zoom into one phrase and point at each note: Snap to the peak reads its frequency, and the time axis gives its length.
Those are the details bird guides describe, so you can compare a call with reference recordings, or two recordings with each other. A whistle reads the same way; for the notes of a melody over time, the pitch detector names each one.
Hum, hiss and noise on an audio spectrogram
Mains hum is a line at 50 Hz or 60 Hz, depending on the country, with its harmonics at 100 and 150 Hz or 120 and 180 Hz above it: a ladder of evenly spaced lines. Hiss is a haze across the top, a fan or a mower a broad band with lines of its own. Point at a hum line on the spectrogram to read its exact frequency.
To check a microphone first, the mic test listens and gives a verdict. To clean a recording, the background noise remover uses AI and costs credits; open its result here and compare the two spectrograms.
What the spectrogram reads: dBFS and the recording
Levels here are dBFS: relative to the loudest a file can hold, so a full-scale sine reads 0 dBFS. Sound pressure in a room, dB SPL, is the job of a calibrated sound level meter, and a microphone’s own response is measured with a reference microphone.
The audio spectrogram shows the recording as captured, the microphone included. For test signals of your own, the tone generator makes tones and sweeps to download and open here.
Who uses an audio spectrogram
Anyone who wants to see a sound rather than guess at it, from a mix to a bird call.

Music and mixing
See the kick, the bass, the voice and the cymbals in one spectrogram, and use the audio spectrum analyzer view to find muddy lows or harsh highs before you reach for an equalizer.

Speech and phonetics
Read vowels and formants in a wideband spectrogram and each harmonic in a narrowband one, with the Gaussian window used in phonetics.

Birdsong
Point at each call on the spectrogram to read its frequency and length, then compare calls and recordings by their numbers. The thrush here is an AI sound effect.

Lossless check
Open a WAV or FLAC and an MP3 of it: the spectrogram shows where the encoder stopped.

Hidden pictures and messages
Find the images and words hidden in songs, games and puzzles with a linear spectrogram up to the top.

Hum and noise
Spot mains hum at 50 or 60 Hz, hiss and fans on the spectrogram, and measure them before you clean a recording.
Spectrogram FAQ
Open the spectrogram
Drop in a file or start the microphone, and read every line of your audio spectrogram: frequency, note and dBFS, free in your browser.



