MP3 vs FLAC

MP3 encodes by discarding: a psychoacoustic model decides which parts of the signal a listener is unlikely to notice, and those parts are removed. What is left is described compactly. The discarded audio is gone, and no amount of converting the file onwards will bring it back.

FLAC compresses without discarding anything. It is the same idea as ZIP applied to audio, tuned for the way sound data repeats — decode a FLAC and you get back the original samples, bit for bit identical to the CD or the master. Typically it lands at half to two-thirds of the uncompressed size, where a 320 kbps MP3 is nearer a fifth.

Usually not, and the page would be dishonest to pretend otherwise. In blind listening tests, a well-encoded MP3 at 256 or 320 kbps is very difficult to distinguish from lossless — most listeners, on most material, in most conditions, cannot do it reliably.

Where the gap opens is at the edges: cymbals and applause, which the encoder handles badly; very quiet passages; good headphones in a quiet room; and lower bitrates, where 128 kbps is genuinely audible to almost anyone once they know what to listen for. On a phone on a train, through earbuds, with traffic outside, the difference is theoretical.

Not audibility — what happens next. Every time a lossy file is re-encoded, the encoder discards more, on top of what has already gone. Convert an MP3 to AAC for a phone, then to Opus for a messaging app, and each step compounds the loss. From FLAC, every one of those conversions starts from the original.

That is the argument that survives scrutiny. You do not know what format you will want in ten years — MP3 replaced nothing gracefully, AAC replaced MP3, Opus is replacing AAC — and a lossless archive can produce all of them. A library of MP3s can only produce worse MP3s.

An album is roughly 300 MB as FLAC against 100 MB as a 320 kbps MP3. A thousand albums is 300 gigabytes rather than 100 — a difference that mattered enormously when it meant a second hard drive and matters much less now that it is a fraction of one.

On a phone it still matters, and that is the case where MP3 or AAC genuinely wins. The arrangement most people converge on is FLAC on the drive or the server and lossy on the device, generated from the FLAC and regenerated whenever the phone or the format changes.

MP3 plays on everything ever made, including a car stereo from 2003 and a USB stick in a hotel television. Its patents expired in 2017, so it is now genuinely free to use, which removed the last argument against it.

FLAC is supported by every serious music player, by Android, by most modern car systems and by home audio equipment, and by Apple everywhere except the one place it matters: iOS has decoded FLAC since version 11 and Safari plays it, but the Music library still refuses to import one and wants ALAC instead. That single gap is why FLAC feels less universal than it is. It is also worth noting that FLAC is an open format with a reference implementation, which is exactly what you want in an archive.

FLAC to MP3 is a normal operation with a real choice attached: use a high bitrate, 256 or 320, or a variable-rate setting at the top of its range. The result is a file that is very hard to distinguish from the source and a fifth of the size, which is exactly what a phone wants.

MP3 to FLAC is the one to avoid. It produces a much larger file containing exactly the same degraded audio — the lossless container preserves what is there, and what is there has already had the high frequencies removed. It looks like an upgrade in a file listing and is not one. The only legitimate reason is a device that will not play MP3 at all.

If you are ripping CDs or buying downloads, take FLAC. It costs storage that is now cheap and it is the version everything else can be generated from.

If you already have an MP3 collection, leave it alone. Converting it to FLAC gains nothing and costs space; re-ripping is only worth it for the records you actually care about. And keep the checksum or the original discs if you have them — the point of an archive is that it can produce the next format, and that only works if it is genuinely the original.

The encoder offers levels from 0 to 8. They change how hard it works to find a compact representation, and they change the encoding time considerably — but every level decodes to bit-identical audio, because the format is lossless throughout.

The practical spread between the fastest and the slowest is only a few per cent of file size, which makes the default of 5 a sensible place to stop. It is worth knowing because the setting looks exactly like a quality control and is not one: turning it up cannot improve the sound, and turning it down cannot damage it.