What Is a Spectrogram? How to Read One (With Examples)

Sound as a picture

A spectrogram turns audio into an image: time runs left to right, frequency runs bottom to top, and colour shows how much energy each frequency carries at each moment. A bass note is a bright band near the bottom; a cymbal crash is a burst reaching the top; silence is a dark column.

Once audio is a picture, questions that are hard to answer by ear become easy to answer by eye — especially “where exactly do these two recordings differ?“

Reading the three axes

Horizontal (time) — each column is a slice of the recording, typically a few milliseconds wide.
Vertical (frequency) — low pitches at the bottom, high pitches at the top.
Colour (energy) — brighter means louder at that frequency and moment.

What makes a mel spectrogram different

Human hearing doesn’t treat all frequencies equally — we distinguish 100 Hz from 200 Hz easily, but 10,000 Hz from 10,100 Hz not at all. The mel scale re-spaces the frequency axis to match how ears actually work, so a mel spectrogram spends its resolution where human perception does. That’s why it’s the standard input for speech recognition — and for perceptual audio comparison.

The spectrogram difference: finding where two files diverge

Comparing two recordings by playing them back to back is slow and unreliable. Comparing their spectrograms side by side is better. Best of all is a difference map: subtract one spectrogram from the other and colour the result, so identical regions stay dark and every divergence lights up.

DiffALL does this automatically. Upload two audio files and you get:

A mel spectrogram of each file.
A difference overlay where mismatches glow — you can see at a glance whether the change is a missing high-frequency band (re-compression), a burst (a click or dropout), or a whole section.
A per-second similarity chart that pinpoints the exact second the files diverge, plus similarity scores based on MFCC, spectral centroid, and RMS energy.

Common things a spectrogram diff reveals

Lossy re-encoding — the high-frequency band is visibly shaved off.
Edits — an inserted or removed section shifts everything after it.
Dropouts and clicks — narrow vertical stripes that don’t exist in the reference.
Different takes — the overall shape matches but details drift everywhere.

Try it

Upload any two audio files — MP3, WAV, FLAC, AAC, OGG, M4A, Opus, or WMA — and read the difference straight off the spectrograms. Free, in the browser, no install.