Lossy decode is one-way: once the bits are gone you cannot turn the quality back up.
How it works
Encoder splits audio into short time-frequency blocks (MP3 ~1152 samples per frame).
A psychoacoustic model finds sounds masked by louder neighbours or below the hearing threshold.
Those masked and inaudible parts get fewer bits or are deleted entirely.
Remaining data is quantised and packed to hit your target bitrate.
Decoder rebuilds an approximation - close to the ear, but never the exact original waveform.
Real examples
Spotify streams Ogg Vorbis/AAC up to ~320 kbps lossy, not the studio master.
A WhatsApp voice note uses Opus near 16-24 kbps - tiny file, obviously squashed.
A YouTube track ripped to MP3 then re-uploaded sounds fizzier each round (generation loss).
Bluetooth playback re-compresses with SBC/AAC/aptX, so even a lossless source gets squashed over the air.
DAB+ digital radio uses lossy AAC, which is why some stations sound 'thin'.
How it helps in live sound
For show playback use WAV/FLAC or 320 kbps MP3 / 256 kbps+ AAC minimum.
Never relay a 128 kbps MP3 through a big PA - the fizzy top end gets ruthlessly exposed.
Get the highest-quality source you can; you cannot fix lost detail with EQ on the night.
Avoid Bluetooth for critical feeds - it re-compresses; use cable, USB, or Dante instead.
Keep one lossless master archive; export lossy copies only for convenience, never as your only copy.
Label files with bitrate so you don't accidentally cue a low-rate web rip on a paying gig.
Everyday analogy
Like saving a photo smaller to email it, then zooming back in, the lost sharpness is gone for good and never returns.
Watch out
Myth: '320 kbps MP3 is CD-quality.' No - it's lossy and discards data permanently; it only sounds close and can never decode back to the true original.
Fun fact
The MP3 psychoacoustic model was tuned partly on Suzanne Vega's a cappella 'Tom's Diner', earning her the nickname 'the Mother of the MP3'.
Key takeaways
Lossy deletes audio permanently - no undo, no rebuild.
It works by binning sounds your ears can't hear (masking).
Higher bitrate = bigger file = closer to the original.
Re-saving lossy stacks damage (generation loss) and sounds fizzy.
For live work use lossless or high-bitrate; quality lost on stage stays lost.