Every track in your library carries a stack of numbers — 44.1 kHz, 24-bit, 320 kbps, 8A, 128.00. Some of them are measurements. Some are guesses. And at least one of them is routinely a lie. Three tools below let you draw the numbers, hear them, and check them against your own library.
01 — The premise
This is the single most useful thing to understand about metadata: almost none of it is verified. A filename, an ID3 tag, a "24-bit / 96 kHz Studio Master" label — these are strings someone typed, or that software wrote on someone's behalf. Nothing in the file format requires them to be true, and nothing checks them when the file changes hands.
A nutrition label is printed by whoever made the food. Weighing it yourself is a different act entirely. Almost everything in an audio file is the label — and this article is mostly about learning which few things are the scale.
Some fields are structural — the decoder genuinely cannot play the file without knowing the sample rate, so that number is real. Others are editorial. Others still are the output of a detection algorithm that was doing its best. Knowing which category each field falls into is most of the skill.
Structural fields are enforced by the decoder and can be trusted. Everything else is an assertion someone made.
02 — The first confusion
The file extension tells you about the box, not the contents. The
container (.wav, .m4a, .ogg) is a wrapper
defining how bytes are laid out and where the tags live. The codec (LPCM, ALAC, AAC,
Opus) is what actually decides whether audio was thrown away.
The container is the envelope; the codec is whether the letter inside was photocopied. Two identical envelopes can hold an original and a fax, and the envelope will never tell you which.
This is why .m4a is genuinely ambiguous: it can hold ALAC,
which is bit-perfect lossless, or AAC, which is lossy. Same extension,
same icon, wildly different files. An .m4a in your library is not
information until you know its codec.
The extension names the box. Only the codec tells you whether anything was thrown away.
03 — The real numbers
By the Nyquist–Shannon theorem, the highest frequency a file can represent is exactly half its sample rate. That single ratio explains the whole spec sheet: 44.1 kHz tops out at 22.05 kHz, which sits just above the limit of human hearing — which is precisely why CD audio chose it. 96 kHz reaches 48 kHz, deep into ultrasonic territory that no listener will hear directly.
Think of a camera filming a spinning wheel. Shoot fast enough and the wheel turns the way it really turns. Shoot too slowly and the spokes appear to crawl backwards — the camera hasn't captured a slower wheel, it has captured the wrong wheel entirely, and no amount of editing recovers the real one. Sample rate is the shutter speed, and the wagon-wheel effect has an audio name: aliasing.
—
Bit depth is amplitude resolution: the number of discrete levels available for each sample. Every additional bit adds about 6.02 dB of dynamic range, pushing quantization noise further beneath the music. 16-bit gives 65,536 steps and a noise floor around 96.3 dB down; 24-bit gives 16.7 million steps and roughly 144.5 dB.
Bit depth buys headroom while you work, not detail you can hear.
03 — The real numbersNote what this means: bit depth buys you headroom while working, not audible detail on playback. It matters enormously in a DAW, where you are stacking and re-gaining dozens of times. On a finished master going to a club system, 16-bit was never the weak link.
Bit depth is a staircase you have to stand on. The music is a smooth ramp; the file can only record which step you were nearest. More steps, closer fit. The gap between the ramp and the step is error — and error you can hear is hiss. Every extra bit doubles the steps and pushes that hiss about 6 dB further beneath the music.
The slider stops at 10 bits because past that the staircase and the ramp are the same line at this scale. Real audio starts at 16 — 65,536 steps, hiss 96 dB down, which is why you have never heard it on a finished record.
For uncompressed audio this is pure arithmetic:
Fs × bit depth × channels. A 16-bit / 44.1 kHz stereo file is
44,100 × 16 × 2 = 1,411.2 kbps, every time. For lossy formats the
bitrate is a budget the encoder was given — and the lower it is, the more the encoder
had to discard.
04 — The landscape
Three categories, and the boundary that matters is between rows 2 and 3 — that is where audio stops being recoverable.
| Format | Codec | Typical bitrate | Frequency ceiling | Verdict |
|---|---|---|---|---|
| WAV | Linear PCM | 1,411 – 9,216 kbps | Full Nyquist | Studio master |
| AIFF | Linear PCM | 1,411 – 4,608 kbps | Full Nyquist | Studio master |
| FLAC | Free Lossless | 600 – 1,100 kbps VBR | Full Nyquist | Archival |
| ALAC | Apple Lossless | 600 – 1,150 kbps VBR | Full Nyquist | Archival |
| AAC | Advanced Audio Coding | 256 – 320 kbps | 18.5 – 20 kHz gentle | Playback |
| Opus | Opus / CELT | 96 – 160 kbps VBR | Adaptive to 20 kHz | Playback |
| MP3 320 | MPEG-1 Layer III | 320 kbps | 20 – 20.5 kHz brickwall | Playback |
| MP3 128 | MPEG-1 Layer III | 128 kbps | 16 kHz brickwall | Degraded |
Lossless compression is worth dwelling on, because it sounds like a contradiction. FLAC and ALAC shrink a file to roughly 50–60% of raw PCM and decode back bit-for-bit identical to the original. Nothing is approximated. It is compression in the ZIP sense, not the JPEG sense — and it is why "lossless" files having a lower bitrate than WAV is not a red flag.
05 — The tell
To hit a bitrate budget, a lossy encoder discards the content it judges least audible — and the cheapest thing to throw away is the top of the spectrum, where there is little energy and poor human sensitivity. The result is a spectral cutoff: a ceiling above which the file contains literally nothing.
These cutoffs are characteristic enough to fingerprint the encoder. MP3 at 320 kbps leaves a hard brickwall at 20–20.5 kHz. MP3 at 128 kbps cuts at 16 kHz — low enough to audibly smear cymbals and add phasing. AAC rolls off more gently between 18.5 and 20 kHz. Genuine lossless has no cutoff at all; content simply continues to the Nyquist limit.
Horizontal axis 0–22.05 kHz — the Nyquist ceiling of a 44.1 kHz file. The shaded region is what the file still contains. Toggle codecs to compare; the dashed outline holds the lossless reference for contrast.
A tone sweeps from 8 kHz upward. With no ceiling it climbs until it leaves your hearing. Put a codec's cutoff in front of it and the sweep stops dead — that silence is the audio the encoder threw away. Headphones, low volume.
Nothing autoplays and no audio loads until you press a button. If the sweep seems to stop early even on "None", that is your own hearing limit, not the file's — which is exactly why encoders bet on discarding this range.
06 — The lie
Anyone can convert a 128 kbps MP3 into a 24-bit / 96 kHz WAV. The file will be enormous, and every tag will read "Studio Master". But transcoding upward only re-samples audio that is already gone — you cannot recreate discarded harmonics. The metadata is fully rewritten; the audio is not.
You cannot recreate harmonics that were already thrown away.
06 — The lieThis is exactly why the spectral cutoff is so useful: it is the one property that survives the laundering. Open the file in a spectrogram — Spek is free, iZotope RX if you have it — and look at the top of the frequency range:
This is the picture Spek would give you. Frequency runs bottom to top; time runs left to right. Both files below claim 24-bit / 96 kHz in their metadata. Only one of them is.
The tell is the straight horizontal edge with nothing above it. Real recordings taper — they never stop dead on a ruled line. That edge is the encoder's brickwall, and re-saving the file at 96 kHz just paints a bigger empty canvas above it.
A "24-bit / 96 kHz" file whose spectrogram flatlines at 16 kHz is a 128 kbps MP3 wearing a costume. The header describes the container it was poured into, not the audio inside it.
A spectrogram costs nothing and settles it in seconds. Look before you trust a tag.
07 — The arithmetic
Since bitrate is data per second, size is just bitrate × duration.
A six-minute track lands at roughly 63 MB as uncompressed WAV, around 35 MB as FLAC
(~45% smaller, still bit-perfect), about 14 MB as a 320 kbps MP3, and under 6 MB at
128 kbps.
Which makes size a fast, free lie detector. A three-minute "WAV" that occupies 4 MB is not a WAV in any meaningful sense — the arithmetic simply does not permit it. You do not need a spectrogram to catch the most blatant cases; you need to divide.
Bitrate times duration is the whole check. A file far smaller than the arithmetic allows is not the format it claims.
—
08 — The estimated fields
Everything above is structural — encoded in the file, verifiable, not up for debate. BPM and key are a different species entirely: they are produced by analysis software making an educated guess, and they are wrong often enough to matter.
BPM detection fails predictably at half- and double-time, which is why a 140 BPM track sometimes arrives tagged 70. Key detection is harder still — it infers a tonal centre from the audio, and struggles with tracks that modulate, sit ambiguously between relative major and minor, or lean heavily on percussion.
BPM and key are the only fields here produced by guessing.
08 — The estimated fieldsThe key field is further confused by competing notations describing identical
information. Musical notation names the key outright (A♭ minor).
Camelot maps the circle of fifths onto a clock face — 12 numbers,
A for minor and B for major, so 8A is A minor.
Open Key does the same with a different offset and m/d
suffixes.
The point of the wheel notations is that adjacency does the theory for you: keys one step apart on the wheel share nearly all their notes, so they mix cleanly. That is not a DJ invention — it is the circle of fifths with a friendlier interface, which is a story worth its own page.
Ask two people to tap along to the same song and one taps twice as often as the other. Neither is wrong — and a detection algorithm faces exactly that choice, with no one to ask. That is why a 140 BPM track turns up tagged 70.
The same groove, counted two ways. Nothing about the audio changes; only the number the software wrote down.
The kick lands in the same place either way. Accented beats are the counting — that is the entire difference between the two tags.
—