What a spectrogram actually shows
A waveform tells you how loud a track is from moment to moment, but it hides what that loudness is made of. A spectrogram opens the sound up into three dimensions at once: time along the horizontal axis, frequency up the vertical axis, and intensity — how much energy sits at each frequency — encoded as brightness or color. Read it like a music-shaped heat map: the further right you look the later in the track you are, the higher up you look the higher the pitch, and the brighter a point glows the louder that particular frequency is at that particular instant.
The picture is built with a Fast Fourier Transform (FFT). The tool slides a short window across the audio — here roughly a thousand samples at a time — and for each window the FFT works out how much energy falls into each frequency band. Stack those slices side by side and you get a continuous image of the spectrum evolving through the song. A short window gives you sharp timing but blurry pitch; a long window gives you fine pitch resolution but smears fast events in time. That trade-off between time and frequency resolution is fundamental to every spectrogram, which is why transients and sustained tones can look so different on the same image.
Reading the axes and the brightness
Start with the vertical axis. Bass and kick energy live near the bottom, vocals and most instruments occupy the busy middle, and cymbals, sibilance and "air" sit up top. The axis is drawn on a logarithmic scale because human hearing is logarithmic — an octave from 100 Hz to 200 Hz feels the same size as one from 5 kHz to 10 kHz, so the tool gives the low and mid bands the visual room they deserve instead of cramming everything musical into the bottom sliver.
Brightness maps to level, usually on a decibel scale so that quiet detail stays visible next to loud peaks. Dark means near silence at that frequency; a bright, hot color means a lot of energy. Once you internalize that, patterns start to jump out. The most useful shapes to recognize are:
- Vertical lines — a short burst of energy across the whole height of the image, i.e. a drum hit or other transient.
- Horizontal streaks — energy parked at one frequency over time: a held note, a bassline, or a resonance.
- A flat ceiling with black above it — a hard bandwidth limit where all content simply stops.
- Broad dark regions — whole bands with little energy, often the top end of a dull master.
Spotting a bandwidth cutoff and lossy MP3s
One of the fastest wins a spectrogram gives you is catching a lossy file masquerading as a master. A true 44.1 kHz WAV can carry content all the way up to about 20 kHz, so its energy should fade out gradually near the top of the image. Lossy codecs save space by discarding the highest frequencies, and they do it with a hard low-pass filter. On the spectrogram that shows up as a flat, ruler-straight ceiling — often somewhere between 15 and 16 kHz for a typical MP3 — with solid black above it.
If you see that wall, the file was almost certainly encoded from a lossy source at some point, even if the container now says WAV. That matters: re-exporting an MP3 as WAV does not restore the missing highs, and shipping it as your master means paying for full-resolution distribution while delivering compressed audio. The spectrogram is the quickest sanity check that the file on your drive is really full-bandwidth before you send it anywhere.
Harsh resonances and ringing
When a single frequency is much louder than its neighbors and holds that way, it draws a bright horizontal streak across the image. Sometimes that streak is musical — a sustained synth pad or an organ drone. But a thin, persistent line that does not move with the notes is usually a resonance: a room mode, a ringing filter, an over-excited de-esser band, or a piece of gear resonating. These are the whistles and harshness that make a mix fatiguing without you being able to point at a note.
The spectrogram tells you exactly which frequency to reach for. Line the streak up against the labelled Hz axis, then apply a narrow EQ cut there and watch the streak dim on a re-analysis. Because you are targeting one band rather than sweeping blindly, you fix the harshness without dulling the rest of the track.
Transients, punch and smearing
Percussion shows up as vertical lines because a hit releases energy across a wide range of frequencies in a very short time. Sharp, clearly separated verticals signal punchy, well-preserved drums. When those lines look soft, doubled or trailed by a faint ghost to their right, you are usually seeing the fingerprint of heavy compression or a low-bitrate codec — the encoder could not keep up with the fast energy and smeared it in time. That smearing is one reason an over-squashed master loses its snap even when the loudness meter looks healthy.
Dull masters and over-rolled highs
If the whole upper half of the image reads dark while the lows and mids are busy, the track is missing air. That can be an intentional vibe, but more often it is an aggressive low-pass, an overloaded limiter eating the highs, or a mix that never got brightened. The spectrogram won't tell you how loud the track is overall — for that you want a LUFS meter — but it will show you exactly where the energy stops, so you know whether to add a high shelf, back off the limiter, or leave it alone. Pairing a spectrogram with a loudness reading gives you both halves of the tonal picture: how much and where.
Hints of AI generation — read them carefully
People often ask whether a spectrogram can prove a track was AI-generated. The honest answer is no — it can only raise a flag. Some AI music renders show a suspiciously regular, repetitive texture up in the high frequencies, or a hard bandwidth cutoff similar to a lossy MP3, because of how the model synthesizes and up-samples audio. Those cues are worth noticing. But plenty of perfectly human tracks look the same after MP3 encoding, and increasingly capable models produce output with completely natural-looking spectra. Treat any visual "AI tell" as a reason to investigate, never as a verdict. When you need a real answer, run the file through our AI Checker, and if it flags, our AI Cleaner removes the artifacts before you release.
Putting it to work: QC before release
A thirty-second spectrogram pass is one of the cheapest quality-control steps you can add to your workflow. Before you upload a master, glance at the image and ask: does the top end reach up near 20 kHz, or is there a lossy wall? Are the drum transients crisp or smeared? Any bright horizontal streaks that scream resonance? Is the upper half alive or dead? Catching a low-bitrate export or a ringing resonance before it reaches a distributor saves takedowns, re-uploads and embarrassment. From here you can round out your checks with the free BPM & Key finder for tempo and key, confirm loudness with the LUFS meter, and if the track is AI-assisted, verify it with the AI Checker and clean it if needed. When you're ready to clean at scale, see our pricing. Everything on this page runs locally in your browser, so you can inspect a file as many times as you like without ever uploading it.