Convert MP3 to FLAC

FLAC is lossless in a precise sense: whatever samples it is handed, it gives back exactly. It makes no claim about where those samples came from. Hand it audio that an MP3 encoder has already thinned out and it will preserve the thinning perfectly, forever, in a file several times larger.

This is the point where most people expected something else, so it is worth stating without softening: nothing in this conversion, or in any conversion, can reconstruct what an MP3 encoder discarded. That data is not compressed, hidden or approximated. It was deleted at encode time and the only place it still exists is the source the MP3 was made from.

A 192 kbps MP3 costs 1.44 MB a minute. Decoded and re-encoded as FLAC, the same audio costs roughly 5 to 6 MB a minute, because FLAC compresses samples rather than perceptions and can only find the redundancy that is genuinely there. A 60-minute album goes from about 85 MB to something near 350 MB.

Across a library that arithmetic is worth pausing on. Converting 200 MP3 albums produces roughly 70 GB of files carrying exactly the information that 17 GB carried a moment earlier. If the goal was a tidy collection, that is a defensible price; if the goal was better sound, it is 53 GB spent on nothing.

MP3 encoders discard everything above a cutoff frequency, because the highest content costs bitrate and contributes least. The cutoff depends on the bitrate — commonly around 16 kHz at 128 kbps, somewhat higher at 320 — and it produces a clean horizontal edge in a spectrogram with nothing above it.

That edge survives the conversion to FLAC unchanged, which makes a transcoded file trivially identifiable. Open the FLAC in any spectrogram viewer and look at the top of the range: a FLAC from a CD runs to about 22 kHz, and a FLAC made from an MP3 stops flat where the MP3 stopped. This is exactly how music communities identify files that have been passed off as lossless.

MP3 carries ID3 tags, a format that grew by accretion over thirty years and shows it: multiple incompatible versions, text encodings that disagree, and fields whose meaning depends on which program wrote them. FLAC carries named text fields with a straightforward specification, and players read them consistently.

So this is a genuine upgrade, and it is about filing rather than audio. Title, artist, album, track number and year all have clear destinations. Convert one album first and look at it in the player you actually use — composer, disc number and sort fields are where mappings tend to go missing, and correcting them across a finished batch takes minutes in a tag editor.

Embedded artwork lives in an ID3 APIC frame on the MP3 side and in a dedicated picture block on the FLAC side, and both hold an ordinary JPEG or PNG with a type label. The concepts line up well enough that artwork usually survives.

Two things are worth checking on a first album. Files with several images — front cover, back cover, disc — do not always carry all of them across, and very large embedded artwork can be a surprising fraction of a small file. If the collection has cover art you care about, verify it once rather than assuming, because re-embedding artwork across 200 albums is genuinely tedious.

There are three, and all of them are about the library rather than the recording. A collection standardised on one format, where an MP3 among the FLACs breaks a script, a player or a policy. A tool that reads FLAC and nothing else. And a recording whose MP3 is the only surviving copy — a lecture, an interview, a broadcast — that you want stored from now on in the format an archive will accept.

One thing FLAC normally offers, this conversion does not. The stream header has a field for an MD5 of the decoded audio, and a decoder years later uses it to prove that what it reconstructs is what was encoded; the muxer here never sees the decoded samples and writes the sixteen zero bytes that RFC 9639 defines as “not known”. If bit rot is the worry, checksum the file yourself and store the digest beside it.

For music, the original usually still exists somewhere: the CD in a box, a purchase that can be re-downloaded, a streaming service that will sell the lossless version. Ripping the CD takes minutes and produces a real FLAC — one whose spectrogram runs to the top and whose extra 250 MB an album is actually carrying something.

Compare that with the alternative honestly. Ten minutes of looking gives a genuine lossless copy; ten seconds of converting gives a large file that only looks like one. The conversion is the right answer when the source is genuinely gone, and the wrong answer roughly as often as people reach for it.

A FLAC converted from an MP3 is indistinguishable from a real one by filename, extension and file size, and completely distinguishable by spectrogram. Shared collections, archives and trackers treat transcoded files as a serious problem for that reason, and they are usually caught.

If you convert for your own filing reasons, keep the result to yourself, or note the provenance in a tag so that the next person — quite possibly you, in five years — does not mistake the file for a master. Honest labelling costs nothing and prevents an MP3 from quietly becoming the archive copy of a recording that had a better one.

Decoding is done by the browser’s own MP3 decoder, the same one it uses to play audio on any web page. Encoding uses a FLAC encoder where the browser exposes one and a WebAssembly build downloaded to the page where it does not, loaded only for FLAC targets and never on any other page.

Both steps happen inside the tab, so a library conversion costs processor time rather than bandwidth, and nothing is uploaded at any point. The limits are a hundred files per drop and 100 MB per file, which for MP3 sources is generous — 100 MB at 192 kbps is over an hour of audio.