A spectrogram is a picture of sound over time. The horizontal axis is time, the vertical axis is frequency (low notes at the bottom, high at the top), and the brightness or colour of each point shows how loud that frequency is at that moment. Where a waveform only shows how loud the sound is, a spectrogram shows what is sounding: which notes, harmonics and noises, and when.
How a spectrogram is made
The sound is cut into short overlapping slices, typically 20 to 100 milliseconds each. Each slice is run through a Fourier transform, which splits it into the frequencies it contains, and the result is drawn as one thin column. Put the columns side by side and you have the spectrogram. This method is called the short-time Fourier transform (STFT).
The slice length is a trade-off. Long slices separate close frequencies well but blur fast events; short slices show sharp timing but smear the frequencies. That's why two programs can draw slightly different pictures of the same file.
A single slice on its own, drawn as a curve of level against frequency, is what a spectrum analyzer shows. A spectrogram is many spectrum analyzer readings stacked side by side, so you can see the sound change.
The shapes you will see
| What you see | What it is |
|---|---|
| A stack of evenly spaced horizontal lines | A pitched sound (a voice, a note on an instrument). The lowest line is the note you hear (the fundamental); the ones above are its harmonics, at 2×, 3×, 4× its frequency. |
| Lines that bend up and down together | A melody or a sung phrase; vibrato shows as a gentle wave. |
| A thin vertical stripe from bottom to top | A click or a drum hit: very short sounds contain almost every frequency at once. |
| A bright blob at the very bottom | Kick drum or bass. |
| Fuzzy, grainy energy over a wide band | Noise: wind, breath, cymbals, a hissing "s" (sibilance, around 5–10 kHz). |
| A steady line at 50 or 60 Hz (and its multiples) | Mains hum from electrical wiring: 50 Hz in Europe and most of Asia, 60 Hz in the Americas. |
| Bright areas fading slowly to the right after each sound | Reverb or echo. |
| A sharp, flat ceiling across the whole recording | A filter or a lossy encoder (see below). |
A good way to learn is to open the live spectrogram, allow the microphone, and make sounds: hum a note and watch the harmonic stack, whistle (one clean line, almost no harmonics), say "sssss", clap. Nothing is recorded; the picture just scrolls as you go.
Linear or logarithmic axis?
On a logarithmic frequency axis every octave takes the same height, the way we hear pitch. Melodies, bass lines and voices are easy to follow, and the picture looks like a piano roll. On a linear axis every kilohertz takes the same height, which gives most of the picture to the treble: better for checking noise, filters and encoder cut-offs. Switch between them depending on what you are looking for.
The Nyquist frequency: the top of the picture
A digital recording can only contain frequencies up to half its sample rate, the Nyquist frequency. A file sampled at 44.1 kHz (CD quality) can hold up to 22.05 kHz; at 48 kHz, up to 24 kHz. Anything higher in the original sound must be filtered out before recording, or it folds back down as false tones (aliasing).
So the top edge of a spectrogram is set by the file's sample rate, not by the music. A 44.1 kHz file drawn to 22 kHz shows everything it can hold. That ceiling is also why listening tests rarely find an audible benefit in very high sample rates: 22 kHz is already above the upper limit of human hearing, which is about 20 kHz for young listeners and lower with age.
How to spot an MP3 disguised as FLAC or WAV
Lossy encoders save space by removing what people hear least, and the first thing to go is the very top of the spectrum. The result is a hard, flat ceiling that runs across the entire song:
- around 16 kHz: typical of a 128 kbps MP3;
- around 19–20 kHz: typical of a 256–320 kbps MP3 or AAC;
- content fading out naturally toward 22 kHz, with cymbals and air reaching the top: a genuine CD-quality source.
To check a file: open it in the spectrogram, switch to the linear axis and show up to 22 kHz, and look at a loud, bright part such as a chorus with cymbals. If a "lossless" file stops dead at 16 kHz from start to finish, it was made from an MP3; converting a lossy file to FLAC can never restore what was removed (see audio formats explained).
Two cautions: some genuine recordings are mastered with a low-pass filter, and old or lo-fi recordings may simply contain nothing up there. A ceiling that comes and goes with the music is normal; a perfectly flat one across the whole track is the giveaway.
What else spectrograms are used for
- Mixing and mastering: finding a resonance, a masking clash between bass and kick, or harshness in a vocal.
- Cleaning recordings: locating hum, hiss or a phone ringing in the background before using noise reduction.
- Speech and singing: watching vowels (formants), pitch and vibrato; singing teachers use live spectrograms for feedback.
- Nature recording: identifying birds and bats by the shape of their calls.
- Music recognition: audio fingerprinting picks the brightest peaks of a spectrogram to identify a song (see what is audio fingerprinting).
FAQ
What does a spectrogram show?
Which frequencies are present at each moment of a sound, and how loud each one is.
What is the difference between a spectrogram and a waveform?
A waveform shows the overall level over time; a spectrogram also splits that level into frequencies, so you can see notes, harmonics and noise separately.
What is the difference between a spectrogram and a spectrum analyzer?
A spectrum analyzer shows the frequencies of one moment; a spectrogram shows them for every moment, side by side.
Why is my spectrogram empty above 16 kHz?
Most often the file is, or once was, a lossy MP3 or AAC at a lower bitrate. It can also be a recording filtered on purpose or a sound with no high content.