Compress an audio file

What a bitrate buys, and when there is nothing left to save

The bitrate is how many bits of storage each second of sound is allowed, and everything about audio compression follows from it. At 320 kbps an encoder has enough room to keep nearly everything; at 128 it must decide what to discard; at 64 it is discarding a great deal and the decisions become audible — cymbals turn to a hiss, applause turns to static, and reverb tails flutter. The useful shorthand: 192 kbps for music going on a phone, 128 for spoken word, 320 when size is not a concern and you want the safest file.

Compressing an already-compressed file is the case to think about before you press the button. The sound has already been through an encoder that threw away what it judged inaudible; a second pass throws away more, starting from less, and cannot tell the original recording from the first encoder's approximations. Re-encoding a 128 kbps MP3 to 128 kbps saves essentially no space and costs real quality — the worst possible trade. Going from 320 to 192, on the other hand, roughly halves the file for a difference most people cannot hear on most equipment.

Where compression genuinely shines is the first pass from something uncompressed. A WAV runs to about 10 MB a minute; the same recording at 192 kbps runs to about 1.4 MB a minute, and the reduction is around nine tenths for a quality difference that takes careful listening on good headphones to identify. That is the compression worth doing, and it is why the format took over.

Speech has far more headroom than music and is worth treating differently. A lecture, a podcast or an interview at 96 or even 64 kbps in mono is perfectly clear and a fraction of the size — a two-hour recording drops from 170 MB to under 60. Halving the channels alone halves the file, and for a single voice there was never any stereo information to lose.

When you want something else

Variable-bitrate encoding is a better idea than picking one number, and ffmpeg exposes it: ffmpeg -i in.wav -c:a libmp3lame -q:a 2 out.mp3 spends bits on the passages that need them and saves them on the quiet ones, landing near 190 kbps on average with better quality than a fixed 190. -q:a 0 is the top setting, around 245 kbps.

For a library rather than a file, foobar2000 and fre:ac convert thousands of files unattended while keeping tags and artwork, and both are free. And if the target is a modern device rather than an old one, Opus at 96 kbps sounds better than MP3 at twice the size — -c:a libopus -b:a 96k — which is a far larger saving than any amount of careful MP3 tuning.