Where the size of an MP3 comes from
An MP3's sound takes a fixed amount of space per second, set by its bitrate. At 320 kbps a minute is about 2.4 MB, at 128 kbps about 0.96 MB, at 64 kbps about 0.48 MB. So the usual way to make an MP3 smaller is to encode it again at a lower bitrate, and that is what every online MP3 compressor does. Two things that the usual approach ignores can matter more than the bitrate.
The first is the picture. MP3s carry their cover art inside the file, and a cover saved straight from a phone or a scanner can be bigger than the sound. The second is the file's own history: an MP3 that is already at 96 kbps doesn't get smaller when it is "compressed" at 128 kbps. It gets slightly worse and a little bigger. This page checks both before it encodes anything.
What we measured
We ran a 20-second test recording (a tone sweep with clicks every second, so any shift in timing shows) through the compressor in Chrome 154 on 13 October 2026, and read every result back with FFmpeg.
| Starting file | Setting | Result |
|---|---|---|
| 320 kbps, 803 KB | 128 kbps | 321 KB, length exactly 20.000 s, starts on the first sample |
| 320 kbps, 803 KB | Fit under 0.3 MB | 281 KB at 112 kbps, the highest standard bitrate that fits |
| 128 kbps with a 1.31 MB cover photo, 1.63 MB in all | 64 kbps, cover kept | 1.47 MB: halving the bitrate saved only 10% |
| Same file | Cover removed | 321 KB, sound not re-encoded (audio bytes identical) |
| Same file | Cover shrunk to 500 px | 368 KB, sound not re-encoded |
| Loud, heavily limited 320 kbps master | 128 kbps | Turned down 2.9 dB; no samples at full scale after decoding |
The cover rows are the reason for the Cover art option. When only the picture changes, the sound isn't decoded at all: the tags at the start of the file are rewritten and the audio is copied byte for byte. If a file you want to email is mostly picture, that is the lossless way to shrink it.
Choosing a bitrate
- Music to listen to: 128 kbps stereo is the everyday choice; 160 or 192 kbps if you'll play it on good speakers.
- Podcasts, lectures, audiobooks, voice notes: 64 kbps mono. Speech doesn't need stereo, and mono puts the whole bitrate into one channel.
- As small as possible: 32 kbps mono, for speech only. Below about 40 kbps per channel the tool also lowers the sample rate to 32 or 22.05 kHz, as the LAME encoder itself would: at that bitrate there isn't room to keep the highest treble, and trying makes voices sound watery.
If you have a limit to meet instead, such as an email attachment cap or an upload form, choose Fit under a file size and type the limit. The tool subtracts the space the tags and cover art take, divides the rest by the recording's length and picks the highest standard MP3 bitrate that fits, then reports the real size. A megabyte here is 1,000,000 bytes, which also satisfies sites that count 1,048,576. Our guide to why a file can fail a limit it seems to meet explains the difference.
Two problems a second encode can cause
A delayed start. Every MP3 begins with a short run-in from the encoder, which players skip using a note in the file. If a compressor decodes the MP3 without reading that note, the run-in becomes real silence in the new file: 25 milliseconds per generation in our first test, until we fixed it. The compressed file now starts on exactly the same sample as the original.
Crackle on loud parts. Music mastered right up to full scale can overshoot it slightly when it is encoded again, and the overshoot is cut off flat on playback. Each result is decoded and checked; if any samples go over, the volume is lowered just enough and the file encoded again.
To change the format rather than the size, MP3 to WAV goes the other way and WAV to MP3 starts from an uncompressed recording, which always gives a better small MP3 than squeezing an existing one. To cut a recording down to the part you need, use Trim audio, which shortens an MP3 without encoding it again.