MP3, WAV, M4A and FLAC: Choosing an Audio Format

Lossy, lossless, and uncompressed

Audio formats fall into three groups, and knowing which group a format belongs to answers most practical questions about it.

Uncompressed formats store the raw measurements. WAV is the example everyone meets: a list of numbers describing air pressure, sampled tens of thousands of times per second. Nothing is analysed, nothing is discarded, and the file size depends only on duration and settings — a minute of silence takes exactly as much space as a minute of orchestra.

Lossless compressed formats — FLAC, ALAC, and others — analyse the waveform for predictable structure and encode it more efficiently, typically halving the size. Decoding gives back bit-identical audio, so nothing is lost. This is strictly better than WAV for storage; the only reason to prefer WAV is that some audio software still expects it.

Lossy formats — MP3, AAC (inside M4A), Opus, Vorbis — permanently discard audio content in exchange for dramatically smaller files, often a tenth of the original. What they discard is chosen using a model of human hearing, which is why a good lossy encode at a sensible bitrate is genuinely difficult to distinguish from the original.

What lossy compression actually removes

Lossy audio compression is built on psychoacoustics — the study of what people can and cannot hear — rather than on mathematics alone. Two effects do most of the work.

Frequency masking means a loud sound conceals quieter sounds at nearby frequencies. A cymbal crash covers a great deal of quiet detail happening around it, and that detail can be discarded without anyone noticing, because it was inaudible in the first place. Temporal masking is the same idea in time: a loud sound also conceals quieter sounds immediately before and after it, giving the encoder a window in which it can be sloppy.

The encoder splits the audio into frequency bands, calculates how much noise each band can hide behind whatever else is happening, and allocates bits accordingly — spending generously where you would hear the difference and barely at all where you would not.

This explains why lossy compression handles some material far better than others. Dense, loud music masks beautifully. A solo harpsichord, a room full of applause, or a cymbal decaying into silence provides very little for quiet detail to hide behind, which is why those are the classic demonstrations of low-bitrate audio sounding obviously wrong.

Picking a bitrate without wasting bytes or quality

For music in MP3, 192 kbps is comfortable and 256 kbps is transparent for practically all listeners in practically all conditions. 320 kbps is the format ceiling and rarely earns its extra bytes — if you genuinely need more than 256 can give you, what you need is a lossless format, not a higher lossy bitrate.

For speech — podcasts, lectures, interviews, voice memos — 96 to 128 kbps is plenty. Voice occupies a narrow frequency range compared with music, and mono halves the requirement again, so a spoken-word recording at 96 kbps mono can sound entirely clean where music at the same setting would not.

AAC, the codec inside M4A files, is more efficient than MP3 and sounds better at the same bitrate, with the gap widest at lower settings. Opus is better still, particularly for speech at very low bitrates. Neither matches MP3 for sheer compatibility, which remains the reason MP3 refuses to die three decades on.

The mistake that costs you twice

Converting one lossy format to another means decoding audio that has already had detail removed, then compressing it again with a different algorithm. The second encoder cannot distinguish the first encoder’s artifacts from the original recording. It faithfully preserves those artifacts, spending bits on them, while discarding something else to stay within budget.

At sensible bitrates this is a minor effect and not worth agonising over — converting an M4A voice memo to a 128 kbps MP3 so it plays in your car is entirely reasonable. At low bitrates, or across several successive conversions, it compounds audibly.

The rule that follows is simple: always convert from the best source you have, and convert once. If you have the original WAV, encode the MP3 from that rather than from an intermediate. If you are extracting audio from a video, work from the original video file rather than a copy that a messaging service already recompressed. And keep the original — the converted file is an export, not a replacement.

The related misunderstanding runs the other way. Converting an MP3 to WAV does not improve anything. You get a file roughly ten times larger containing exactly the same audio, artifacts included. It is worth doing when software you are using demands uncompressed input, and useless as a quality upgrade.

Which format for which job

Use WAV while you are working. Editing, mixing, transcription, sampling, and analysis all benefit from audio that decodes instantly, survives unlimited re-saving without generational loss, and is accepted by every tool without argument.

Use FLAC for archiving. It is lossless, so nothing is compromised, and it is roughly half the size of WAV with far better metadata support. There is no quality argument for archiving in WAV instead.

Use MP3 for sharing and playback on anything you do not control — car stereos, older players, voicemail systems, upload forms. It is the safest possible choice and the reason MP3 conversion works in every browser on this site: the encoder ships with the page rather than depending on your browser’s capabilities.

Use M4A when you are inside the Apple ecosystem and nothing needs to leave it. It is what Voice Memos and GarageBand produce, it sounds better than MP3 at the same size, and it works flawlessly until the moment you hand it to a device that only speaks MP3 — which is precisely when converting becomes worth doing.