Lossy audio compression achieves large size reductions by permanently discarding parts of the signal a psychoacoustic model predicts the listener won't hear — masked detail and sounds below the threshold of hearing — rather than restoring the original exactly. AAC, Opus, MP3, and the Dolby and MPEG codecs are all lossy, and at sensible bitrates the result is transparent, indistinguishable from the source to most ears. The tradeoff is that the discarded information is gone for good, so repeated lossy re-encoding (transcoding) accumulates artefacts, and lossy files make a poor archival format. Lossy is what makes streaming and real-time audio practical; lossless is what you keep the masters in.
