Reading Audio by Spectrogram: Spotting Fake Lossless and Noise
Hunting audio problems by ear is feeling an elephant in the dark: when the noise starts, where highs roll off, which passage clips — a spectrogram draws all of it as one readable picture.
How to read a spectrogram
Horizontal axis is time, vertical is frequency, brightness is energy: the entire track laid out as one image. Healthy music shows "sculptural" texture — the low end (bottom) persistently dense, highs (top) flickering as instruments enter, generally thick below and sparse above, breathing with the rhythm.
Two basic reads: the "ceiling" and the "floor". A ceiling that cuts off cleanly near 16kHz is the signature of MP3 lossy encoding. A file claiming lossless with a flat ceiling at 22kHz yet showing a lossy cutoff is fake lossless — MP3 data wearing a WAV shell.
Locating three classes of problems
Noise: steady background noise (hum, air conditioning) draws as a horizontal bright line spanning the whole timeline — its height IS its frequency, telling you which denoiser setting applies. Transient noise (keyboards, doors) draws as vertical bright bars, time-stamped to the second — that decides where to cut.
Clipping and silence: clipping shows as abnormal harmonic bands above the signal — over-gained recordings are unmistakable. Long dark regions mark silence or near-silent passages — a "where is there no content" scan for transcription and editing prep, done in one pass.
Use cases and limits
Three high-frequency scenarios: verifying purchased or downloaded "lossless" music (a 16kHz cutoff means fake); recording QC (examine noise shape before denoising, check residue after); problem localization (vague complaints like "noise in part three" become a ten-second lookup of time and frequency).
Know the boundary: spectrograms excel at WHERE the problem is, not whether it sounds acceptable — final judgment belongs to your ears. It is a diagnostic, not a mixing tool: read, listen to confirm, then act.