CloudlessKit ← All guides

CloudlessKit / Guides

How to shrink audio files without ruining the sound

Updated July 22, 2026

An hour of CD-quality stereo audio saved as WAV runs about 600 MB. An hour of spoken-word recording, handled right, can fit in under 15 MB with no real loss to a listener's ear. That gap comes down to four settings, and once you know what each one does, you can stop guessing and start picking numbers on purpose.

What shrinks an audio file

Bitrate is the big one. It's how many kilobits of data get spent per second of sound, written as something like 128 kbps or 32 kbps. Higher bitrate means more data describing the waveform, which usually means better sound and a bigger file. Lower bitrate throws detail away. For music this shows up as a kind of underwater, swishy quality once you go too low. For speech it shows up much later, because there's less detail to lose in the first place.

Codec choice matters almost as much as the number next to it. MP3 is decades old and needs more bits to sound clean than newer codecs doing the same job. Opus was built more recently specifically to sound good at low bitrates, so a 32 kbps Opus file usually sounds noticeably better than a 32 kbps MP3 of the same audio. AAC, the format inside M4A files, sits in between. If you're chasing small size above all, codec choice can matter more than the bitrate slider itself.

Mono versus stereo is the one people forget. A stereo file stores two separate channels, left and right, and at a given bitrate setting that total budget is split between them, so each channel gets less of it than it would if the same bitrate were spent on one channel alone. That split is worth it for music, where the two channels genuinely carry different information. But a huge number of voice recordings, phone interviews, dictaphone recordings, a laptop mic picking up a room, are effectively mono already: both channels carry nearly the same signal because there was only ever one source and one mic. Converting that kind of file to mono lets the whole bitrate go to a single channel instead of splitting it, which is why for voice you can drop to mono and then cut the bitrate roughly in half with no perceived loss: it throws away a near-duplicate channel, not real information.

Sample rate is the last lever, and it's the least understood. It's how many times per second the original sound wave gets measured, usually 44,100 or 48,000 times for music. Human speech carries almost all of its meaning below about 4,000 Hz, so 16,000 samples per second is already more than enough to keep it clear. Music, especially anything with cymbals, strings, or high harmonics, needs the full range. Dropping sample rate on a voice recording is close to free. Dropping it on a song is not.

Voice tolerates a lot more than music

Put those four levers together and the practical gap between voice and music is enormous. A podcast interview, a lecture recording, a deposition, all of these are single-channel content with a narrow frequency range and no listener expecting a wide stereo soundstage. You can push the bitrate way down, drop to mono, and lower the sample rate, and a person listening on earbuds or a phone speaker will not notice.

Try that on a song and the compromises show up fast. Cymbals turn to hiss, bass loses its edges, and the stereo width that makes a mix feel spacious collapses into something flatter. Music generally wants stereo and needs a real bitrate cushion, something in the range of 128 to 192 kbps for MP3, or noticeably less for Opus, before it starts sounding compressed in the bad sense of the word.

This is also why the same file can be a totally different problem depending on what it's for. A student recording a lecture on their phone doesn't need concert-hall fidelity, they need to hear the professor clearly and search the file later. A band bouncing a final mix for a client absolutely does need the fidelity, and shrinking that file the same way you'd shrink a lecture would be a mistake.

A one-hour interview, by the numbers

To make this concrete, here's roughly what one hour of a spoken interview weighs at common settings. Actual results vary a little by codec and encoder, but these are close enough to plan around.

Notice how far that last row is from the first. Same hour of audio, roughly sixty times smaller, and for a plain spoken conversation most people can't tell the difference on a phone or laptop speaker. Below about 12 kbps mono is where things start to sound thin and a little metallic, so that's a reasonable floor to keep in mind before you push a slider all the way down.

When compressing is a mistake

Not every file should be touched. If your source is already an MP3, M4A, or Opus file, and you re-encode it again at a lower bitrate, you're compressing something that's already had detail thrown away once. The artifacts from each pass stack, so a file that's been through two lossy encodes sounds noticeably worse than one encoded straight from the original. If a file is already small and already lossy, check its size before assuming it needs more work.

Archival recordings deserve the opposite instinct. A one-of-a-kind interview, a legal recording, a multitrack music session you might remix later, the original take of anything you can't re-record: keep at least one untouched, full-quality copy before you make a smaller version for sending around. Storage is cheap. A detail you compressed away years ago is not coming back.

One way to check what you're starting with: look at the file extension and, if you can, its properties or "get info" panel. A .wav or .flac file is lossless and safe to compress once. A .mp3, .m4a, .aac, .ogg, or .opus file has already been compressed at least once, whether you did it or someone else did, and squeezing it further has a real cost.

And sometimes the honest answer is to do nothing.

If you're emailing a two-minute voice note to a colleague, or storing something on a drive with plenty of room, there's no size problem to solve. Compression is a tool for when a number somewhere is actually smaller than your file, not a step you owe every audio file by default.

Why people end up here

The reasons this becomes urgent are almost always the same handful of caps. Email is the most common: most webmail providers cap attachments at a few dozen megabytes, so a raw hour-long WAV interview simply won't attach, full stop.

Podcast hosts are the second big one. Free and lower-tier plans commonly cap the size of an individual episode upload, and even without a hard cap, storage adds up fast across a season recorded at an unnecessarily high bitrate. Course platforms and school upload portals do something similar with lecture recordings and assignment submissions, often with caps that feel arbitrary until you've hit one.

If you're recording voice memos on a phone, check what you're starting with before assuming you need to compress at all. Phones already save recordings in a compressed format like AAC, so a five-minute voice memo might already be a couple of megabytes rather than the tens of megabytes raw audio would imply. The real problem cases are usually long ones: hour-plus interviews, multi-hour lecture blocks, or anything captured from a device set to record uncompressed audio.

Working out the exact bitrate to hit a size limit by hand is tedious, since it depends on the file's length and the format's overhead. Audio Compressor handles that math directly: give it a target size like "under 25 MB" and it works backward to the bitrate that gets you there, or pick mono and a lower bitrate yourself if you'd rather set it manually. Either way, none of it requires uploading the file anywhere first.

If what you actually need out of a long recording is a searchable transcript rather than a smaller audio file, that's a different problem with a different tool: Audio Transcriber turns the recording into text instead, which is often more useful than a smaller file when the goal was never to listen to it again.