What MP3 Actually Throws Away (and When WAV Still Matters)

A WAV file isn't a compression format at all — it's a thin container around raw PCM (pulse-code modulation) audio, the direct digital representation of the sound wave. Every sample is stored at full precision, which is why the file sizes are so predictable: they're pure arithmetic.

The math behind WAV's size

Uncompressed audio size is sample rate × bit depth × channels. CD-quality audio is 44,100 samples per second, 16 bits per sample, 2 channels:

44,100 × 16 × 2 = 1,411,200 bits per second ≈ 1,411 kbps

That works out to about 10.6 MB per minute, or roughly 635 MB for an hour of stereo audio. A 320 kbps MP3 of the same material is about 4.4x smaller; a 128 kbps MP3 is 11x smaller. Nothing about the WAV is "higher quality" in some mystical sense — it simply keeps all the data, and the size follows directly from that.

What lossy encoders actually remove

MP3 and AAC don't shrink audio the way ZIP shrinks a file. ZIP-style (lossless) compression finds redundancy and can reverse itself perfectly; raw audio has too little exploitable redundancy for that to get you much. Lossy codecs instead run a psychoacoustic model — a model of human hearing — and delete signal content the model predicts is inaudible:

• Quiet sounds below the hearing threshold at each frequency are dropped outright.
• Frequency masking: a loud tone makes nearby quieter frequencies inaudible for a moment, so the encoder stores those masked frequencies with far less precision, or not at all.
• Temporal masking: a loud transient briefly masks quieter sounds just after (and even slightly before) it.
• At lower bitrates, most encoders also apply a lowpass filter, cutting the highest frequencies entirely to spend the bit budget where it's more audible.

The key word is delete. Decoding an MP3 back to WAV reconstructs a signal that sounds close to the original, but the discarded detail is gone permanently — the WAV you get out is a full-size container around an already-reduced signal.

Why 128 and 320 kbps sound different

The bitrate is the encoder's budget. At 320 kbps there's enough room that the model only discards things that are genuinely very hard to hear — for most listeners on most equipment, most material is indistinguishable from the source. At 128 kbps the encoder must cut far more aggressively, and the failures become audible on difficult material: cymbals and hi-hats smear into a swirly wash, sharp transients pick up a faint pre-echo, and the top of the frequency range dulls. Dense, bright, percussive music exposes 128 kbps quickly; a solo podcast voice may sound fine far lower.

The transcoding trap

Lossy-to-lossy conversion compounds the damage. Converting a 128 kbps MP3 to AAC means decoding it (artifacts and all), then running a second psychoacoustic model over that already-degraded signal, which discards a different set of detail. Each generation stacks its own artifacts on the last. And the reverse never works: re-encoding a 128 kbps MP3 at 320 kbps, or exporting it as WAV, just wraps the same degraded audio in a bigger file. When you convert with an audio converter, the quality ceiling is always the worst lossy step anywhere in the file's history.

When WAV is worth it — and when it's waste

WAV earns its size in exactly two situations. Editing and processing: every effect, mix, or trim followed by a lossy export is one more generation of loss, so working files should stay uncompressed (or losslessly compressed, like FLAC) until the final export. Archival: a lossless master lets you encode to whatever format matters in ten years without inheriting today's codec artifacts.

For everything else — listening, sharing, uploading — WAV is mostly wasted bandwidth. A well-encoded high-bitrate lossy file is audibly transparent for the overwhelming majority of listeners, at a tenth of the size. If a recording only needs to be heard, running it through an audio size reducer costs you data you were never going to hear; if it's ever going to be edited again, keep the lossless original, because that's the one conversion you can't undo.