What Is Audio Compression? File Size vs Dynamic Range

"Audio compression" is one phrase for two unrelated jobs. One makes an audio file smaller so it is easier to store, send or stream. The other is an effect in music and podcast production that evens out how loud different moments are. They share a name and nothing else, and mixing them up sends people to the wrong tool. This guide explains both: how each one works, what the numbers mean, and which one you actually need.

The two meanings of audio compression

File-size compression (also called data compression) encodes audio so it takes up less space. MP3, AAC, Opus, FLAC and ALAC all do this. The result is a smaller file that plays back as the same sound, or very nearly the same.

Dynamic range compression is a mixing and mastering effect. It turns down the parts of a recording that go above a level you choose, which shrinks the gap between the loudest and quietest moments. The file stays the same length, the same format and the same size. Only the loudness changes.

One reduces bytes. The other reduces decibels. If you are not sure which one you need, this table should settle it:

Part 1: File-size compression

Uncompressed audio is big. CD-quality sound stores 44,100 samples per second, 16 bits each, for two channels. That is 1,411 kbps: about 10.6 MB per minute, 42 MB for a four-minute song and 635 MB for an hour. File-size compression exists to bring those numbers down. It comes in two kinds.

Lossless compression: smaller, with nothing lost

A lossless format stores the audio in less space and gives back exactly the same samples when you play it. It works by spotting patterns. In most music, each sample is close to the one before it, so the encoder predicts the next sample and stores only how far off the prediction was. Instead of storing 100, 101, 103, 102, 104, it stores 100, +1, +2, −1, +2. Small numbers need fewer bits, and a coding step then packs them tightly.

FLAC typically brings audio down to 50 to 70% of its original size. How much it saves depends on the music: quiet, simple recordings shrink more than loud, dense ones. The main lossless formats are:

• FLAC — open and free, the standard for music libraries and hi-res downloads. See what is FLAC.
• ALAC (Apple Lossless) — Apple's equivalent, used by Apple Music and iTunes.
• WavPack and Monkey's Audio (APE) — older alternatives with smaller followings.

What about ZIP? Zipping an audio file is also lossless, so it never harms the sound. But a general-purpose ZIP is not built for audio. The FLAC project itself compares FLAC to ZIP and notes that FLAC compresses audio much better. Zipping an MP3 or AAC file saves almost nothing, because that data is already compressed.

Lossy compression: much smaller, by removing what you probably will not hear

A lossy format goes further. It uses a model of human hearing, called a psychoacoustic model, to decide which parts of the sound you are unlikely to notice, and it spends few or no bits on them. The main effects it relies on are:

• Frequency masking. A loud tone hides quieter tones close to it in pitch. While a loud 1 kHz tone plays, a soft tone at 1.1 kHz is very hard to hear.
• Temporal masking. A loud sound also hides quieter sounds just before and just after it: for roughly 20 milliseconds before and about 100 milliseconds after.
• The threshold of hearing. Some sounds are simply too quiet to hear at their pitch, so they can go.
• Shared stereo. Much of the left and right channel is the same, so joint stereo stores the shared part once.

Whatever is removed is gone for good. Decoding a lossy file never brings it back, which is why converting an MP3 to WAV makes a bigger file, not a better one.

How an MP3 encoder works, step by step

1. It splits the audio into frames of 1,152 samples, about 26 milliseconds each at 44.1 kHz.
2. It turns each frame from a waveform into its frequencies, using a filter bank and a transform called the MDCT.
3. It runs the psychoacoustic model to find which frequencies are masked in that frame.
4. It shares out the bits available at the chosen bitrate, giving more to what you will hear and less to what is masked.
5. It rounds the frequency values to fit that budget. This rounding is where the loss happens.
6. It packs the result tightly (with Huffman coding) and writes the frame.

Newer codecs such as AAC and Opus follow the same idea with better tools and models, which is why they generally sound better than MP3 at the same bitrate.

The formats you will meet

How much smaller does compression make a file?

Compared with CD-quality audio at 1,411 kbps:

In real terms, 1,000 four-minute songs take about 42 GB as CD-quality WAV, 9.6 GB as 320 kbps MP3 and 3.8 GB at 128 kbps. An hour-long mono podcast at 96 kbps is about 43 MB, against 635 MB as CD-quality stereo WAV.

Transparency: when compression becomes audible

A compressed file is "transparent" when you cannot tell it from the original in a blind test. There is no single bitrate where that happens for everyone: it depends on the codec, the encoder, the music, your equipment and your ears. Many listeners find MP3 at 256 to 320 kbps very hard to tell from CD audio, and codecs such as AAC and Opus reach a similar point at lower bitrates. Our WAV vs MP3 guide shows how to run a blind test on your own headphones.

Below that point, the usual signs are:

• A faint smear just before sharp hits like drums, called pre-echo
• Swishy, smeared cymbals and hi-hats
• A narrower, flatter stereo image
• Voices that sound hollow or watery at very low bitrates

Constant, variable and average bitrate

• CBR (constant bitrate) gives every second the same number of bits. File sizes are predictable, but simple passages get as many bits as busy ones.
• VBR (variable bitrate) spends more bits on complex passages and fewer on simple ones, which usually gives better sound for the same size.
• ABR (average bitrate) is a middle path: it varies, but aims at a set average.

For most music, VBR is a sensible default. Our VBR vs CBR guide compares them in detail.

Do not compress twice

Converting one lossy format into another, such as MP3 to OGG or MP3 to a smaller MP3, stacks the losses: the second encoder removes more, working from a signal that is already damaged. The drop is not always audible, but it is real. When you have the original WAV or FLAC, encode from that.

When file-size compression is worth it

• Sending a recording by email or chat, where attachment limits apply
• Fitting a music library on a phone
• Streaming, which is why Spotify delivers compressed audio instead of WAV
• Publishing a podcast, where listeners download every episode

For any of these, drop your file into the Audio Compressor, or use a format-specific page: Compress MP3, Compress WAV, Compress M4A, Compress OGG or Compress FLAC. They run in your browser, so your files are not uploaded.

Part 2: Dynamic range compression

Now the other meaning. Open Audacity or any music software, look through the effects, and you will find a "compressor". This compressor does not make the file smaller. It changes how loud the parts of the audio are compared with each other.

What a compressor does

A compressor watches the level of the sound. When the sound rises above a threshold you set, it turns the sound down by an amount set by the ratio. At 4:1, every 4 dB above the threshold comes out as 1 dB above it, so a peak 8 dB over the threshold ends up only 2 dB over. Anything below the threshold passes through untouched. Because the loud parts are now quieter, you then raise the whole track with makeup gain. The result: quiet parts end up relatively louder, loud parts relatively quieter, and the whole thing sounds more even.

The five main controls

1. Threshold (in dB) — the level where compression starts.
2. Ratio — how hard the compressor turns down sound above the threshold. 2:1 is gentle, 4:1 is moderate and 10:1 is heavy.
3. Attack (in milliseconds) — how quickly it reacts once the sound crosses the threshold. A fast attack catches the start of a drum hit; a slower one lets that first punch through.
4. Release (in milliseconds) — how quickly it lets go once the sound drops back below the threshold.
5. Makeup gain (in dB) — the boost applied afterwards to bring the level back up.

Many compressors also have a knee setting, which makes compression start gradually around the threshold instead of all at once.

Compressor vs limiter

A limiter is a compressor pushed to the extreme: a very high ratio (10:1 and above, up to "infinity to one") with a fast attack. Its job is not to shape the sound but to stop peaks from going over a set ceiling. Mastering engineers often put a limiter last in the chain for that reason.

Common starting points

These are typical places to start, not rules. Adjust by ear, since every voice and recording is different.

• Spoken voice or podcast: ratio around 3:1 to 4:1, medium attack, medium release, then a limiter to catch stray peaks.
• Singing: ratio around 3:1 to 4:1, set so only the louder phrases are turned down.
• Drums: ratio around 4:1 with a slower attack, so the hit comes through before the compressor clamps down.
• Whole mix ("glue"): a gentle 1.5:1 to 2:1, touching only the loudest moments.

How to hear compression

Compression is easiest to hear when it is overdone. Listen for:

• Pumping — the background level rising and falling in time with loud notes
• Lost punch — drums that sound flat because their attack has been clamped
• Breathing and room noise that becomes louder in the gaps between phrases
• Harsh "s" sounds that stand out more once the rest of the voice is turned down

Switch the compressor on and off at the same perceived volume. If the "on" version only sounds better because it is louder, raise the level on the "off" version and compare again.

Why dynamic range compression exists

Real recordings can swing a long way between whispers and loud hits. Earbuds on a train or a car stereo on the highway cannot make the quiet parts audible without the loud parts becoming too loud. Compression narrows that gap, so the whole recording can sit at a comfortable level.

It also explains the "loudness war", when many masters were compressed ever harder to sound louder than the competition. Streaming services have taken away much of the reward for that. Spotify, for example, adjusts tracks to about −14 LUFS by default (−11 on its Loud setting and −19 on Quiet), so a heavily squashed master is simply turned down to match everything else.

Doing it with AudioUtils

AudioUtils does not have a compressor with threshold and ratio controls. Dynamic range compression is a creative job that needs listening and adjusting, and it belongs in an audio editor: Audacity is free and has a Compressor effect, and music software such as Logic, Reaper or Ableton offers more.

What we do have is the Audio Normalizer. It sets a file's overall loudness to −14 LUFS with peaks kept under −1 dBTP, which is useful when several files play at noticeably different volumes. It evens out loudness between files, not the ups and downs within one take. For that, use a compressor. Our guide to audio normalization explains the difference.

Which one do you need?

If a file is too big, you need file-size compression. Choose lossless (FLAC) to keep every detail, or lossy (MP3, AAC, Opus) for much smaller files, and start from the best original you have.

If a recording sounds uneven, you need dynamic range compression in an audio editor, perhaps followed by loudness normalization so it matches your other files.

And if you need both, which is common for podcasts, do them in that order: compress the dynamics while editing, then export a compressed file at the end.