Blog · Methodology

How FFT spectrum analysis exposes audio quality

What the analysis actually measures, why bitrate tags can lie, and how to turn frequency content into a usable 0–100 score for an entire library.

Download Free → See pricing

The problem

Most people know FFT (Fast Fourier Transform) as the math behind audio visualizers. It's also the math that catches fake-320 MP3s and fake FLACs — but the connection between "colorful waveform display" and "forensic audio quality verification" isn't obvious. The bitrate tag in an MP3 is metadata, set by whoever encoded it; the actual audio quality is in the signal, and FFT is how you read the signal.

How Music Library Doctor does it

1 FFT converts audio from time-based samples to frequency-based components. A 1-second slice of audio gets broken down into the strength of every frequency component — bass at 60 Hz, midrange at 1 kHz, cymbals at 12 kHz, air at 16 kHz, etc.
2 Real high-quality audio has energy across the full spectrum (up to ~20–22 kHz depending on sample rate). Lossy compression — MP3, AAC, OGG — discards high-frequency content first because that's what listeners notice least. A 128 kbps encoder typically cuts off everything above 15–16 kHz; a 192 kbps encoder, around 18 kHz; a 320 kbps encoder preserves nearly the full range.
3 FFT shows the cliff. If you analyze a fake-320 MP3, you see a hard drop in spectral energy right where the original lossy encoder stopped. A real 320 keeps going. A FLAC made from a 128 kbps source has the same 16 kHz cliff as the 128 kbps source.
4 Turning that into a score means defining a quality model: where the cliff lands, how steep it is, what the energy distribution looks like below the cliff. Music Library Doctor's quality scorer combines those signals into a 0–100 number per file.
5 Library-scale auditing means running this analysis on every file, fast. The FFT itself is cheap; the bottleneck is decoding lossy audio formats to PCM. MLD pipelines decode + analysis so a 10,000-track library finishes in roughly an hour, then results are cached for instant re-runs.

Supported today

Rekordbox · Serato DJ · VirtualDJ (incl. Favorite Folders) on Windows 10+ and macOS (Apple Silicon + Intel).

Why native integration matters

FFT-based scoring is reliable because it measures what's physically in the file, not what the tag claims. The tradeoffs are: it can't catch some artifacts (compression smearing, stereo image collapse) that aren't visible in the magnitude spectrum, and it can flag legitimately low-bandwidth audio (vintage recordings, voice-only) as "low quality" when it's actually just narrow-band by nature. Music Library Doctor's scorer applies sanity checks for those cases — vintage recordings and spoken word get marked as "low-bandwidth source" rather than fake. The 0–100 score plus the secondary indicators give you enough to decide per file or per cluster.

Frequently asked questions

Why doesn't every music app do this?

Most apps are designed for playback, not auditing. Adding FFT analysis would slow down library scans without delivering value most users would notice. MLD's bet is that DJs and audiophiles do notice, and would pay $19–49 for the answer.

Can FFT analysis be fooled?

It can be edge-cased. Someone could encode a fake-320 with synthetic high-frequency noise above the original cliff to mask the cliff. That's expensive in artifacts and rare in practice. For the typical fake-320 (just a re-encode), the cliff is obvious.

Does this require special hardware?

No. Modern CPUs do FFT in microseconds. The bottleneck is decoding the lossy audio to PCM — a 5-minute MP3 takes a few seconds to decode. CPU-bound, not memory-bound.

Is the analysis stored locally?

Yes. MLD caches scores per file in its local SQLCipher-encrypted database. Subsequent scans use the cache, only re-analyzing files that changed.