MP3 vs WAV: we encoded the same audio six ways and weighed it

A WAV holds every sample the recording captured. An MP3 throws most of the data away and sounds almost the same. That trade is the entire question, and it is easier to settle with numbers than adjectives, so here is one 30 second stereo recording at 44.1 kHz put through five encoders on 2026-09-13:

A WAV is not compressed at all, which is why its size is pure arithmetic: 44,100 samples a second, two channels, two bytes each, is 10.6 MB a minute whatever the recording contains. Silence and an orchestra cost exactly the same.

What lossless means, tested rather than asserted

FLAC is called lossless, so decoding it back should return the file it came from. It does. The FLAC above, decoded to WAV and compared against the original with cmp: byte identical. Not close, not indistinguishable. The same file. That is a third of the size with nothing given up, and it is the part most comparisons skip.

What lossy means, in the same test

The MP3 at 320 kbps, the highest setting most software offers, decoded back and compared sample by sample against the original:

• 99.0% of samples came back different.
• The worst one was off by 217 out of 32,768, which is 0.66% of the full scale.

Both numbers matter and most articles give you only one. “Nearly every sample is altered” sounds alarming and is true. “The worst error is two thirds of one percent” is why nobody hears it. MP3 does not remove the music, it stops describing it precisely, and it spends its effort on the parts of the signal a person is least able to detect.

The consequence is not how it sounds today. It is that the discarded detail is gone. Convert that MP3 to WAV and you get a large file containing exactly the MP3's data, not the original recording. Nothing recovers it.

So which one

WAV while you are recording and editing, and for a master you might process again. It is also what audio software, games and hardware expect when they want something with no decoding cost.

FLAC for archiving. It is the same audio as the WAV at roughly a third of the size, and the test above is the reason to trust that sentence. The only catch is that fewer things open it than open an MP3.

MP3 when it has to work everywhere without being thought about. It is the oldest of these and the only one you can hand to anyone, on anything, with no question about support. 128 kbps for speech, 192 or 320 for music you will listen to properly.

AAC, the usual contents of an .m4a, sounds better than MP3 at the same bitrate and is what Apple and YouTube use. Pick it when the destination is a phone or a modern app rather than an unknown.

Opus is the best of them per byte, which the 96 kbps row shows: a fifth of the WAV and smaller than MP3 at 128. It is the default for voice chat and a good choice for speech. It is also the least widely supported, so it is the wrong answer for a file you are sending to a stranger.

The one conversion that is never worth doing

MP3 to WAV, hoping to improve it. The file gets about eleven times larger and contains no more information than it did, because the detail was discarded at encode time. If something demands a WAV, that is a fine reason to convert. Quality is not.

filetity is built by Adarsh Mishra. Every figure above came from encoding one 30 second stereo file with ffmpeg on 2026-09-13; your audio will give different sizes, and the ordering will not. Corrections welcome: support@filetity.com.