Why Your Audio Tags Disappear When You Convert (And When That Is a Feature)

You convert an album from FLAC to MP3, load it into your music player, and every track shows up as "Unknown Artist". The audio is perfect. The names, the album, the cover art - all gone.

That is deliberate on our part, and it is worth explaining properly, because the reasoning is genuinely interesting and the trade-off is real. For most music libraries it is an inconvenience with a straightforward fix. For a surprising number of other files, it is the whole point.

What is actually inside an audio file

An audio file is two things bolted together: the encoded sound, and a block of metadata describing it. In MP3 that block is called an ID3 tag; FLAC and OGG use Vorbis comments; M4A and other MP4-family files store it in atoms. Different names, same idea.

Everyone knows the standard fields - title, artist, album, year, track number, genre, comment. What is much less widely known is that these formats are extensible. ID3 in particular allows arbitrary user-defined text fields, which means any software touching the file can add whatever it likes, under whatever name it likes, and every player will simply carry it along.

So in practice audio files pick things up. Recording apps write device models and software versions. Digital audio workstations add project names and session identifiers. Ripping tools stamp their own name and the date. Corporate transcription and dictation systems attach reference numbers and user accounts. And the comment field, being free text, ends up holding whatever somebody typed.

Here is a real example of what a single MP3 can carry, all of it perfectly valid and invisible in a normal music player:

• title - Interview
• comment - client: Acme Corp, do not share
• recording_device - Zoom H6
• location - 51.5074,-0.1278
• operator - [email protected]

Three of those five are custom fields that no standard requires and no player will show you. A file like that looks completely ordinary while carrying an email address, a set of coordinates and a confidentiality note.

And then there is the cover art. An embedded image is a complete JPEG or PNG living inside the audio file - which means it brings its own metadata with it. A photo used as artwork can carry the EXIF data any photo carries, camera model and timestamps included.

Why we remove all of it

Faced with that, a converter has three options.

It could copy everything through, which is what most tools do. Convenient, and it silently forwards every hidden field into the new file - which is fine for an album and not fine for a recording you are about to send somebody.

It could copy a safe subset - title, artist, album, and nothing else. This sounds like the sensible middle path and it is worse than it looks, because it requires deciding on your behalf which fields matter. It also fails quietly and confidently: a field the allowlist did not anticipate gets dropped without a word, and a field it did anticipate gets forwarded even when it should not have been.

Or it could remove everything and be clear about it. That is what we do. Every conversion is instructed to discard all input metadata, and cover art, chapters and any attached streams are dropped alongside it. What you get back is the audio and nothing else.

The one exception, for completeness: the output carries a short encoder tag naming the software that wrote the file. It comes from our encoder, not from your input, and reveals nothing about the source.

The reasoning is the same principle behind stripping EXIF from photos. Metadata is the part of a file people forget exists, so the safe default is to remove it, and it is far easier to add a tag back than to discover months later that you shared one you did not know was there.

When this is exactly what you want

Music is not the only thing people convert. A large share of audio conversion is voice, and voice recordings are frequently sensitive in ways music never is:

• Journalists converting interview recordings before sending them to an editor or a transcription service.
• Lawyers and investigators handling recordings where a device identifier or a stray comment could matter.
• Clinicians and researchers converting session recordings that must not carry identifying details.
• Anyone sharing a voice memo, which arrives with whatever the phone's recording app decided to attach.
• Musicians sending demos, where the exported file often still holds the project name, the studio and the DAW.

In every one of these cases, "the tags are gone" is not a limitation. It is the reason to use the tool.

What it costs you with a music library

Being honest about the downside: if you are converting an album, or a few hundred, you will get files with no artist, no album, no track numbers and no artwork. Your music player will not organise them, and fixing that by hand would be miserable.

So do not fix it by hand. Tagging tools solve this properly:

• MusicBrainz Picard (free, on Windows, macOS and Linux) identifies tracks from the audio itself using acoustic fingerprinting, then writes correct tags and cover art from a community database. It does not need your existing tags, which is precisely why it works here. Point it at the converted folder and most of a library sorts itself out.
• Mp3tag (Windows, with a macOS version) is excellent for bulk edits and for rebuilding tags from filenames, if your files are named like 01 - Artist - Title.mp3.
• Your music player. Several, including recent Apple Music and foobar2000, can look up and fill in track information for untagged files.

The practical order for a library conversion is: convert everything, then run Picard over the output folder once, then check the handful of albums it could not identify. It is one extra step in a job that already takes an afternoon.

One useful habit: keep your original files. They hold the tags you already had, so if a tagging tool struggles with an obscure album you can always read the correct values off the source.

A note on what is not removed

Removing tags does not make a recording anonymous, and it would be misleading to suggest otherwise. The audio still contains whatever was said, in a recognisable voice, in a room with its own acoustic character. Metadata stripping addresses the accidental disclosure - the fields you did not know were attached - not the content.

Nor does it affect quality. Tags sit alongside the audio stream, not inside it, so removing them changes nothing about how the file sounds. It usually makes it slightly smaller, since embedded artwork is often a few hundred kilobytes of the file size.

How this fits the rest of the conversion

Metadata stripping is one part of a consistent approach. Every audio conversion runs inside a sandbox with no network access at all, under a memory cap and a hard time limit, working entirely in server memory rather than on a disk. The no-network property is not decoration: media files can reference external resources, and a decoder with no route out cannot fetch them.

Your file is never stored. It exists for the seconds the conversion takes and is discarded when the download is sent. There is no retention period, no account holding a history, and nothing to delete afterwards - which is a reasonable standard for a file that might be an interview, and a perfectly good one for a file that is just an album.