How do you compress an MP3, and how much smaller can it get?

Why bitrate is the only real dial

An MP3 at a constant bitrate spends the same number of bits on every second, so its size is the bitrate times the length, whatever the music. That is why silence and music made files of the same size in our merge tests, and why a mono file is no smaller than a stereo one at the same bitrate.

So making an MP3 smaller always means decoding it and encoding it again with fewer bits. Mono and a lower sample rate do not shrink the file by themselves; they let the encoder spend a lower bitrate on less sound (one side instead of two, a narrower range of frequencies), which is why the WAV to MP3 page offers bitrates below 128 kbps only once you choose one of them. Each step is lossy, and re-encoding the same MP3 again and again adds up.

How we tested

• Music: 3 minutes (1:00 to 4:00) of Mozart's Symphony No. 40, first movement, played by the Musopen Symphony. It is in the public domain on Wikimedia Commons as a 48 kHz, 24-bit FLAC. We converted it to 44.1 kHz, 16-bit stereo; that WAV is the master.
• The MP3s to compress: the master encoded with LAME 3.100 (through FFmpeg 4.4.2) at a constant 320, 192 and 128 kbps, with title, artist and album tags, the way many downloaded and ripped MP3s are made.
• Browser: Chromium 153 driven by Playwright, opening a local copy of this site's audio cutter, MP3 to WAV and WAV to MP3 pages. One fresh tab per file; whole file, no fades, volume 100%.
• Score: signal-to-error ratio in dB against the master after lining the files up. Higher is closer; it measures how far the waveform moved, not what you hear. Mono files are scored against the master's mono mix, so the score shows what the encoder lost, not the stereo spread (see below).
• Top frequency: the highest frequency where the file's average spectrum stays within 20 dB of the master's. The master itself reaches 21.3 kHz.

Size and score for each setting

All rows start from the same 320 kbps MP3 (7,202,598 bytes for 3 minutes). The first two rows are the original MP3s from LAME, for comparison.

At the same size, mono beat stereo every time: 96 kbps mono scored 23.5 dB against 15.4 dB for 96 kbps stereo at 32 kHz, and 64 kbps mono 17.0 dB against 14.0 to 14.3 dB for 64 kbps stereo. With a single signal instead of two, the encoder has twice the bits for it.

The share column is just the bitrate ratio: 128/320 is 40%, 32/320 is 10%. The 44.1 kHz stereo 128 kbps file came out the same from both routes (15.2 dB, 16.0 kHz), so the cutter is the shorter way to get it.

What mono and a lower sample rate take away

• Mono throws away the stereo spread. Mono mixes left and right into a single signal, so whatever differs between them is lost. In this orchestral recording, the difference between left and right carried 25.4% of the energy. Against the stereo master, every mono file scored only 5.5 to 5.9 dB, and so did a perfect mono mix of the master (5.9 dB): that loss comes from going mono, not from the encoder. A recording that is already mono, or where both sides are the same, has nothing to lose here.
• A lower sample rate cuts the high frequencies. At 32 kHz nothing was left above 15.0 kHz, at 24 kHz above 10.9 kHz, at 22.05 kHz above 10.0 kHz (8.3 kHz at 32 kbps), and at 16 kHz above 7.6 kHz.
• The length stays the same. Every file played 180.04 to 180.07 seconds against the master's 180.00; the extra is encoder padding at the end, as the WAV to MP3 page's FAQ says.

Starting from a compressed file, and re-saving at the same bitrate

Your MP3 is already lossy, so we checked how much that costs on the second encode.

• From 320 kbps, the second encode cost nothing measurable. The scores matched encoding the uncompressed master directly. From 192 kbps they were 0.2 to 1.0 dB lower.
• Re-saving a 128 kbps MP3 at 128 kbps is a loss for nothing. The new file was 2,880,574 bytes against 2,881,109 (only the tags and header went), and the score fell from 21.5 to 15.3 dB.
• This site's encoder is weaker than LAME 3.100 at the same bitrate. The site uses lamejs, a JavaScript port of LAME, with a fixed bitrate. LAME's own 128 kbps file scored 21.5 dB; the site's 128 kbps scored 15.2 dB from every source, and 27.7 against LAME's 28.0 dB at 192 kbps. So if your MP3 is already 128 kbps, saving it at 128 kbps here makes it worse, and you need 96 kbps mono or lower to gain size.

Tags and time

• Tags: the audio cutter's MP3s had no tags. Through MP3 to WAV and WAV to MP3, title, artist and album came through to the small MP3 (both pages said "kept title, artist, album"); the encoder field written by LAME did not.
• The WAV in between is big: 31,752,254 bytes for 3 minutes (the page shows 30.3 MB). You can delete it after the second step.
• Time for 3 minutes: MP3 to WAV took 2.2 s. The MP3 export took 23.7 to 25.2 s at 128 kbps in the cutter and 24.7 s in WAV to MP3, but only 7.0 to 8.0 s for the mono 22.05 and 16 kHz files at 32 and 48 kbps. Measured on a server, not a phone.

Steps

Down to 128 kbps (from 192 kbps or more):

1. Open the audio cutter and choose your MP3.
2. Type 0 as the start and a large number, such as 99999, as the end. Leave both fades at 0.
3. Pick MP3 and 128 kbps, and press Download.

Smaller than 128 kbps, or to keep the tags:

1. In MP3 to WAV, choose the MP3, leave 16-bit and Same as the MP3, press Convert and download the WAV.
2. In WAV to MP3, pick the sample rate and Mono first; the bitrate list then shows the low bitrates. For music, 96 kbps mono at 44.1 kHz kept the most at 30% of the size. When size matters most, 48 kbps mono at 22.05 kHz gave 15% with sound up to 10 kHz.
3. Choose the WAV, press Convert and download the MP3. Listen to it before you delete anything.

Questions

Does the audio cutter offer anything below 128 kbps?

No. Its MP3 export offers 128, 192, 256 and 320 kbps. The lower bitrates, mono and lower sample rates are on the WAV to MP3 page.

Can I shorten the file instead?

Yes, and it saves size without touching the sound: half the length is half the size at any bitrate. Cut out what you don't need in the audio cutter, or split a long recording into parts.

Is my file uploaded to be compressed?

No. All three pages decode and encode the file in your browser; see how to check whether an online tool uploads your file.