Reference

Audio glossary

Audio discussion is full of jargon, and a lot of it gets used badly — sometimes by people selling you something. These are the terms that come up around bitrates, codecs and lossless audio, defined the way I'd explain them to someone who just wants to know if their files are any good.

A

AAC (Advanced Audio Coding)

The lossy codec that succeeded MP3, and genuinely better at the same bitrate. It's what Apple Music, iTunes and YouTube use. Usually found inside an .m4a container, occasionally as a bare .aac file. Because it's more efficient than MP3, its frequency cutoffs sit higher for a given bitrate — judge it by AAC's own numbers, not MP3's.

ABR (Average Bitrate)

A middle ground between CBR and VBR. The encoder varies bitrate with the music but steers toward a target average, so you get a predictable file size with some of VBR's efficiency.

ABX test

The only listening test worth arguing about. You hear sample A, sample B, then X — secretly one of the two — and you say which. No labels, no prices, no expectations. It's the standard way to establish whether a difference is actually audible, and it humbles nearly everyone who tries it. See can you tell the difference.

AIFF

Apple's uncompressed PCM format. Functionally the same thing as WAV with a different header, and vulnerable to all the same abuses.

ALAC (Apple Lossless)

Apple's answer to FLAC. Genuinely lossless, compresses to roughly half the size of WAV, and lives inside an .m4a container — which is why you can't tell an ALAC from an AAC by the file extension alone.

B

Bit depth

How many bits describe each individual sample. It controls dynamic range — roughly 6 dB per bit — so 16-bit gives about 96 dB and 24-bit about 144 dB in theory. It does not control frequency response; that's sample rate. Converting 16-bit to 24-bit adds eight bits of zeros and no information whatsoever.

Bitrate

How many bits the file spends per second of audio, in kbps. It's a measure of spending, not quality — a 320 kbps file made from a 128 kbps file is still 128 kbps audio in an expensive wrapper. Uncompressed CD audio is 1,411 kbps; FLAC of the same material lands around 700–1,000; a good MP3 is 192–320.

C

CBR (Constant Bitrate)

Every second of audio gets the same number of bits, whether it's a cymbal crash or silence. Predictable and mildly wasteful, but the reported bitrate is always honest — which is more than you can say for badly-headered VBR.

Codec

Short for coder-decoder: the scheme that turns audio into data and back. MP3, AAC, FLAC, ALAC and Opus are codecs. Not to be confused with the container, which is just the box the codec's output sits in.

Container

The file wrapper that holds encoded audio plus metadata. .m4a, .ogg and .wav are containers; the codec inside can vary. This distinction is the source of endless confusion — and of the entire fake-lossless problem, since a lossless container will happily hold lossy audio.

D

DAC (Digital-to-Analog Converter)

The chip that turns your file into a voltage your headphones can use. Every phone and laptop has one. Upgrading it matters far less than upgrading your headphones — see the hardware guide.

dBFS (Decibels Full Scale)

Loudness measured against the maximum a digital system can represent. 0 dBFS is the ceiling; everything else is negative. A 16-bit file's noise floor sits around −96 dBFS, which is how you catch a 24-bit file that was padded rather than genuinely recorded at 24-bit.

Dither

A tiny amount of deliberate noise added when reducing bit depth, which converts ugly correlated distortion into benign hiss. Relevant here for a side effect: dither noise runs all the way up to the Nyquist limit, so a genuine CD rip is never completely empty at the top of the spectrogram. A transcode is.

Dynamic range

The distance between the quietest and loudest parts of a recording. Determined technically by bit depth and artistically by whoever mastered it — and the second factor matters far more than the first.

F

FFT (Fast Fourier Transform)

The algorithm that breaks a chunk of audio into its component frequencies. It's what turns a waveform into a spectrum, and it's the engine behind every spectrogram and every bitrate estimate on this site.

FLAC (Free Lossless Audio Codec)

The standard open lossless format. Compresses to roughly 50–70% of WAV with zero loss — decode it and you get the original samples back exactly. Also the most commonly faked format in existence, because nothing stops anyone wrapping an MP3 in it. How to check yours.

G

Generation loss

Degradation that accumulates each time lossy audio is re-encoded. Like photocopying a photocopy. Lossless formats have none of it, which is the main practical argument for archiving in FLAC even if you listen to MP3s.

H

HE-AAC

High Efficiency AAC, built for very low bitrates (32–64 kbps) using SBR. Notable because it's the one common case where a spectrogram looks full but the high frequencies are synthesized rather than real. Mostly found in broadcast streams and podcasts.

Hi-res audio

Marketing term for anything above CD quality — typically 24-bit and/or 88.2 kHz and up. Sometimes it's a genuinely better master. Sometimes it's a CD that was upsampled and sold at a premium. The spectrum tells you which.

J

Joint stereo

An encoding trick that stores what the two channels share once, instead of twice. Saves a lot of bits on most music. It got a bad reputation from crude early implementations; modern encoders handle it well and you should generally leave it on.

K

Killer sample

A recording that reliably defeats lossy encoders. Harpsichord, glockenspiel, castanets, applause — sounds with sharp transients or dense noise. Codec developers use them as torture tests, and they're where the difference between lossy and lossless is easiest to actually hear.

L

LAME

The best MP3 encoder there is, and the one behind most decent MP3s made in the last twenty years. Its V0 preset is the sensible default for VBR. It stamps a header into files it creates, which is occasionally useful evidence — an MP3 encoder's signature inside a "lossless" file is not a good sign.

Lossless

Compression that throws away nothing — decode it and you get a bit-perfect copy of the source, like unzipping a document. FLAC, ALAC, WavPack. Crucially, this describes the format, not the audio's history: a lossless file made from an MP3 is losslessly storing damaged audio. See what is a truly lossless file.

Lossy

Compression that permanently deletes audio information judged inaudible — mostly high frequencies and sounds masked by louder ones. MP3, AAC, Opus, Vorbis. Modern lossy codecs at high bitrates are extremely good. The data is still gone forever.

Lowpass filter

A filter that lets low frequencies through and blocks high ones. Every lossy encoder applies one, because high frequencies are expensive and hard to hear. Where it sits is a direct function of bitrate — and it's the fingerprint this entire site is built on reading.

M

M4A

An MPEG-4 audio container. It can hold lossy AAC or lossless ALAC, and the extension gives you no clue which. Always check the codec before judging an .m4a.

N

Nyquist frequency

Half the sample rate, and the absolute highest frequency a digital file can represent. For CD audio at 44.1 kHz that's 22.05 kHz; at 96 kHz it's 48 kHz. This is the ceiling every measurement on this site is compared against — a genuine file gets close to it, a transcode doesn't come near.

O

Opus

The best lossy codec currently in general use, especially at low bitrates. Open, royalty-free, and what YouTube, Discord and most voice apps reach for. An Opus file at 128 kbps comfortably beats an MP3 at the same size.

P

PCM (Pulse-Code Modulation)

Raw uncompressed digital audio — just a long list of sample values. It's what's inside a WAV or AIFF, and what every codec decodes back to before it reaches your speakers.

Peak frequency ratio

A number this site reports: the highest frequency actually measured in a file, divided by its Nyquist limit. Close to 1.0 means the file is using its full available range, as a genuine lossless file should. Around 0.7 means roughly a third of the available spectrum is empty, and you should ask why.

Pre-echo

A lossy artifact where a sharp transient smears backwards in time, producing a faint ghost just before the hit. Most audible on castanets, harpsichord and close-mic'd percussion. A classic sign of a low-bitrate encode.

Psychoacoustics

The study of how hearing actually works, including its blind spots — quiet sounds hidden by louder ones, poor discrimination at frequency extremes. Every lossy codec is essentially a catalogue of these blind spots, exploited to save bits.

S

Sample rate

How many times per second the waveform is measured, in Hz or kHz. CD is 44.1 kHz, video is usually 48 kHz, studio work often 96 kHz. It determines the highest representable frequency (see Nyquist) and nothing else — a higher sample rate cannot improve audio that was captured at a lower one.

SBR (Spectral Band Replication)

The trick behind HE-AAC: encode only the lower spectrum, then have the decoder fabricate the high band by copying and reshaping it. Efficient, and the one thing that can make a very low-bitrate file look full on a spectrogram. Look for a texture seam at the crossover, usually 8–12 kHz.

Spectrogram

A picture of frequency content over time: time across the bottom, frequency up the side, brightness for energy. The fastest way to spot a fake — a lossy cutoff appears as a hard horizontal line with dead space above it. Once you've seen one you can't unsee it.

T

Transcode

Converting from one encoded format to another — and in practice, the word people use for the specific sin of converting lossy audio into a lossless format and calling it lossless. A transcode can never recover what the first encoder deleted; it can only carry the damage forward in a larger file.

Transparent

Indistinguishable from the source under blind testing. For most listeners on most material, 320 kbps MP3 and 256 kbps AAC are transparent. Note that this is a claim about perception, not about data — a transparent file is still missing information, you just can't hear which.

U

Upsampling / upscaling

Converting a file to a higher sample rate or bit depth than it was captured at. The file gets bigger; the information does not increase. Easy to detect — an upsampled 96 kHz file has nothing above 22.05 kHz, leaving the top half of the spectrogram conspicuously empty. How to spot it.

V

VBR (Variable Bitrate)

The encoder spends bits where the music needs them, dropping in quiet passages and spiking in dense ones. Better quality per megabyte than CBR. The catch: there's no single bitrate to report, so files missing a proper Xing header get misreported by players reading only the first frame.

Vorbis

An open lossy codec, usually found in .ogg files. Competitive with AAC in its day, now largely superseded by Opus from the same developers. Still turns up in games and older open-source projects.

W

WAV

Uncompressed PCM in a Microsoft container. No compression, no quality of its own, and no clue about provenance — a WAV made from a 96 kbps MP3 is a very large file full of 96 kbps audio. Because size is fixed by sample rate and bit depth, file size tells you nothing at all. Checking a WAV.

X

Xing header

A small block at the start of a VBR MP3 recording the frame count, which lets players calculate the true average bitrate and seek accurately. When it's missing — common in files that have been split or badly re-tagged — players guess from the first frame, and a perfectly good VBR file can appear to be 96 kbps.

Something missing, or defined badly? Tell me and I'll fix it.