Audio compressor: make the file smaller

An audio compressor for file size, not for dynamics

Two unrelated things are called an audio compressor, and only one of them is here. This one makes the file take fewer bytes. The other one is a mixing tool that evens out the distance between the loud parts and the quiet parts of a recording, and it usually makes the file slightly larger rather than smaller. They share a word by accident. If you came looking for the mixing tool, nothing on this page will help you; if you came because an upload limit or a mailbox rejected a recording, you are in the right place.

What happens here is a re-encode. The audio is decoded to raw samples and handed to the LAME encoder at a bitrate you pick, and the bitrate is very nearly the whole story: a minute of 128 kbps MP3 is about 960 kilobytes almost regardless of what the minute contains. That is why an estimate can be shown before the work rather than after it, and why the estimate is arithmetic instead of a guess.

Every setting is a trade, and each one says what it costs

Bitrate is the main lever. Four common answers sit on buttons, and the encoder's own sixteen rates sit in a list beside them, because a default should be a shortcut and never a ceiling. For speech, 64 kbps is usually indistinguishable from the original at a quarter of the size of the usual default; for music, 96 kbps is about as low as most people accept.

The other two settings are not free size reductions and are not presented as such. Mono halves what the encoder has to spend, but it is one channel where there were two: the pair is averaged rather than picked, so a guitar panned hard left moves to the middle instead of vanishing. Lowering the sample rate cuts the top off the frequency range before the encoder ever sees the audio, which is a real change to the content — though at a very low bitrate it can sound better anyway, because the encoder stops spending bits on detail it could not have represented well.

A compressor that hands back a bigger file has failed

Re-encoding an MP3 that is already smaller than your chosen bitrate produces a larger file that sounds no better, having now been through two generations of loss. Most tools will hand you that file anyway, and the name of the tool invites you to assume you gained something. So when the output is not smaller, no download is offered here at all. The receipt says how much larger it came out, says that your original is the smaller file, and tells you to pick a lower bitrate. The measured bytes in and the measured bytes out are both on that receipt either way, not an estimate repeated back to you.

The finished file is also reopened and decoded before it is offered, so the duration, channel count and sample rate on the receipt are read out of the actual output rather than out of the settings that were requested. An encoder that quietly produced a forty-second file from a forty-one-second source would be caught by that check rather than by you, later.

What it reads, and where the engines disagree

Anything the browser can decode can go in: MP3, WAV, FLAC, Ogg, AAC, M4A and AIFF are all offered. Browser codecs are not uniform, though, and pretending otherwise would just move the failure later — so the line under the file picker says which formats depend on your browser, and the file is opened and a first stretch decoded before any setting is offered. FLAC is the live example: Chrome and Edge decode it, while Safari's engine can refuse it, and you learn that before the conversion rather than after.

If your source is a WAV and you simply want the ordinary 128 kbps result, WAV to MP3 is the shorter road. To change format rather than size, the audio converter handles the other directions, and MP3 to WAV goes back to uncompressed PCM. Often the real answer is less audio rather than smaller audio: trim it to the part you need, or drop the silence, and the file shrinks in proportion without a second generation of loss. For a video file whose soundtrack is not the problem, the video compressor is the right address.

Nothing leaves the tab, and nothing is refused for its size

The decoder and the encoder both run inside this page. No part of the file is uploaded, no request carries it anywhere, and neither its contents nor its name is written to a log or to the address bar — you can watch the network panel for the whole run. A large file is not turned away either: the note about what a long recording will likely cost in memory arrives before the work begins rather than once the tab is already struggling, and then it runs. Memory, not patience, is what actually runs out, and a tab that runs out of it reloads without explaining why.