How to Reveal Hidden Images and Messages in a Spectrogram

Why audio can contain pictures

A spectrogram maps time and frequency onto two visible dimensions. A creator can synthesize tones so that their energy traces letters, symbols, or an image when displayed. This technique appears in experimental music, game easter eggs, alternate reality games, puzzles, and spectrogram art.

Not every face-like or letter-like shape is intentional. Compression blocks, harmonics, stereo effects, and random noise can create suggestive patterns. Treat the first sighting as a clue, not proof.

Use settings that preserve geometry

Start with a linear frequency scale. Equal vertical distances then represent equal frequency differences, which is how many spectrogram images are designed. Try FFT 4096, then 8192 if edges look soft or narrow lines merge.

Switch to a monochrome or high-contrast color map. Increase contrast slowly and adjust brightness until the background recedes without erasing faint edges. The Reveal Hidden Image preset is a fast starting point, but manual refinement is often necessary.

• Linear frequency scale
• FFT 4096 or 8192
• Low smoothing for crisp edges
• Monochrome or high-contrast color map

Search the full recording methodically

Watch the moving five-second window during playback and note suspicious times. Encoded content may be very short, placed near silence, or concentrated above the range emphasized by normal listening. A full spectrogram export makes it easier to scan the complete file before returning to a close window.

Check left, right, and mixed channels when available. Some puzzles hide content in one channel or rely on differences between them. Avoid repeatedly transcoding the file because lossy compression can blur thin lines.

Verify before decoding

A real encoded image should remain recognizable across modest changes in FFT size, brightness, and color map. Look for deliberate alignment, repeated stroke thickness, or structure that matches the timing of the audio. Compare against nearby sections that contain similar noise.

Export the current PNG for close inspection and CSV when you need the underlying time-frequency values. Preserve the original audio and record your settings so another person can reproduce the result.