How MP3 Works
MP3 achieves its small file sizes through perceptual coding: instead of storing the audio waveform faithfully like WAV does, it throws away the parts of the sound a human ear is unlikely to notice and keeps only what matters perceptually. The decision about what to discard is driven by a psychoacoustic model, a mathematical description of the limits of human hearing built from decades of listening research.
Two effects of human hearing do most of the work. The first is the absolute threshold of hearing: we simply cannot hear very quiet sounds, and that threshold varies with frequency (we are far more sensitive around 2 to 5 kHz than at the extremes). Anything below the threshold can be removed entirely. The second, and more powerful, is auditory masking: a loud tone makes nearby quieter tones inaudible, both in frequency (a loud note hides a soft one close to it in pitch) and in time (a loud sound briefly hides quiet sounds just before and after it). MP3 measures where these masking "shadows" fall and allocates almost no data to the sounds hiding inside them.
Mechanically, the encoder runs the signal through a hybrid filter bank that splits it into frequency sub-bands, applies the psychoacoustic model to decide how precisely each band must be stored, and then quantizes each band so that the unavoidable rounding error (quantization noise) is pushed down beneath the masking threshold where the ear will not hear it. The result is stored as a stream of short, self-contained frames, each with its own header and roughly 26 milliseconds of audio. Because every frame stands alone, an MP3 can be streamed, cut, or joined without re-encoding the whole file, which is a large part of why the format took over the early internet.
How aggressively the encoder discards data is set by the bitrate. A constant bitrate (CBR) spends the same number of bits per second throughout; a variable bitrate (VBR) spends more on complex passages like a busy chorus and less on simple ones like silence or a solo voice, giving better quality for the same average size. As a practical guide, 128 kbps is the old "good enough" web standard but audibly imperfect on good equipment, 192 kbps is a comfortable everyday quality, and 320 kbps (the maximum) is near-transparent for most listeners. Below about 96 kbps the artifacts, a swirly, underwater quality on cymbals and a smearing of sharp transients, become obvious.
History and Standardization
MP3 grew out of research led by Karlheinz Brandenburg and colleagues at Germany's Fraunhofer Institute through the 1980s, building on academic work in audio compression and psychoacoustics. The format was formally standardized as Layer III of the MPEG-1 audio specification (ISO/IEC 11172-3) in 1992 to 1993; "MP3" is shorthand for MPEG-1 Audio Layer III, and the familiar .mp3 file extension was settled on in 1995. A later revision under MPEG-2 (ISO/IEC 13818-3) added support for lower sampling rates and bitrates, useful for speech and low-bandwidth applications.
A piece of audio folklore from the development is genuinely instructive: Brandenburg used Suzanne Vega's a cappella recording of "Tom's Diner" as a reference track precisely because a solo human voice is unforgiving, the ear knows exactly what a voice should sound like, so any compression artifact stands out. Tuning the encoder until that track survived intact is part of why MP3 handles voice so well.
MP3's explosion in popularity in the late 1990s was inseparable from the early internet: files small enough to download over dial-up, combined with portable players and file-sharing networks, made it the format that turned digital music into a mass phenomenon. For years, though, MP3 was patent-encumbered, with Fraunhofer and its partners collecting licensing fees from encoder and decoder makers. Those core patents expired by 2017, and Fraunhofer's licensing program ended, effectively making MP3 a royalty-free format that anyone can implement freely, one reason it remains so universally supported.
MP3 Compared to Newer Codecs
By any purely technical measure, MP3 has been surpassed. AAC, the codec used by Apple Music, YouTube, and most streaming services, generally sounds better than MP3 at the same bitrate, especially on stereo imaging and at lower bitrates. Opus, the modern open codec, beats both at low and medium bitrates and is the default for web voice and many streaming applications. If you are choosing a lossy format today purely on quality per byte, MP3 is not the winner.
What keeps MP3 relevant is compatibility, not fidelity. Nearly every device ever built that plays digital audio, old car stereos, cheap players, embedded systems, every operating system and browser, can play an MP3 without question, and now that it is royalty-free there is no licensing friction to using it. For a file you need to be certain will play anywhere, MP3 is still the safest choice.
One limitation is permanent and worth understanding: MP3 is lossy, so the data its psychoacoustic model discards is gone for good. You cannot recover the original quality by converting an MP3 back to WAV or FLAC, that only produces a larger file wrapped around the already-reduced audio, and re-encoding an MP3 to another lossy format compounds the loss. For this reason, editing and archival workflows keep a lossless master (WAV or FLAC) and treat MP3 strictly as a delivery format generated at the very end.
