Audio formats, and what the bitrate number means
The number after the format name tells you how much of the sound was kept. Knowing which number to pick saves you from files that are either ruined or pointlessly large.
Audio formats split into two families, and the first decision is always which family you need. Everything else is detail.
Lossless keeps everything
WAV and FLAC store every sample the microphone or the mix produced. WAV stores them raw, at about 10 MB per minute for CD-quality stereo. FLAC compresses them the way a zip file would, typically to half or two-thirds of the WAV size, and decodes back to the identical samples. There is no quality setting because there is no quality lost. You use these when the audio is going to be edited, processed or kept as the master, because every other format below throws information away and cannot get it back.
Lossy keeps what you can hear
MP3, AAC (usually in an M4A or MP4 file), Opus and Vorbis (usually in an OGG file) all work by modelling human hearing and discarding the parts of the signal the model says you will not notice: quiet sounds masked by loud ones nearby, frequencies above what most adults can hear, stereo detail in bass that the ear cannot place. Done at a reasonable bitrate, this is genuinely inaudible to nearly everyone. Done aggressively, it produces the swishy, underwater artefacts you associate with early internet audio.
The four are not equal. They are, roughly, in order of age and of efficiency:
| Codec | Transparent for music at about | Good for speech at about | Where it plays |
|---|---|---|---|
| MP3 | 192 to 256 kbps | 96 to 128 kbps | Everywhere. Every device made in the last 25 years. |
| AAC | 128 to 192 kbps | 64 to 96 kbps | Everywhere modern. The default in MP4 video and on Apple devices. |
| Vorbis | 128 to 192 kbps | 64 to 96 kbps | Browsers, games, Linux. Largely replaced by Opus. |
| Opus | 96 to 128 kbps | 24 to 48 kbps | Browsers, WhatsApp, Discord, WebM. Not Apple's music apps. |
"Transparent" means listeners cannot reliably tell it from the original in a blind test. Opus at 96 kbps sounds as good as MP3 at 192, which is why voice chat and streaming services moved to it. MP3's advantage is purely that everything plays it, and that advantage is real: if the file is going to somebody else and you do not know what they will play it on, MP3 remains the safe answer, and the converter defaults to it.
What the bitrate number is
Kilobits per second: how much data describes each second of sound. At 128 kbps a minute of audio is about 0.96 MB; at 320 kbps about 2.4 MB. The number is a budget, and a higher budget lets the encoder keep more. But it is a budget for the encoder, not a measure of the source. This is where most mistakes happen.
You cannot add quality back
If a file is already lossy, converting it to a higher bitrate does nothing except make it larger. A 128 kbps MP3 converted to 320 kbps is a 320 kbps file containing 128 kbps of information, with a little extra damage from having been decoded and re-encoded. Converting it to WAV does the same thing with ten times the size: a lossless container around lossy contents.
The same applies to extracting audio from video. The sound inside an MP4 is almost always AAC at somewhere between 96 and 192 kbps. Pulling it out to a 320 kbps MP3 gains nothing over 192, and pulling it out to M4A without re-encoding at all, which the extract tool offers, loses nothing and takes a second. Use media info to see what is actually in the file before deciding.
Generation loss
Every lossy encode discards something. Encoding something that was already encoded discards more, and the artefacts compound. One conversion from a lossless master is inaudible at a sensible bitrate. Three conversions, each from the last, are not. If you have to process a file (trim it, change its volume, merge it with another), do all of the processing in one step, from the best source you have, and encode once at the end. The tools here each run FFmpeg once, so chaining three of them means three encodes; if you find yourself doing that, the terminal and a single command is better, and the command panel on each tool shows you the pieces.
Sample rate and channels
Two other numbers appear on audio files. The sample rate, 44.1 kHz for CDs and music, 48 kHz for video and broadcast, is how many times per second the sound was measured. Converting between the two is harmless but unnecessary unless a target demands one; video editors generally want 48 kHz. Channels: stereo for music, mono for a single voice. A podcast recorded with one microphone and exported in stereo is carrying two identical copies of itself. Mono at 64 kbps Opus, or 96 kbps AAC, is a perfectly good spoken-word file and a quarter the size of what most people export.
Volume
Loudness is separate from format, but it comes up in the same breath. A recording that is too quiet can be raised with the volume tool; a recording that is too loud and clipping cannot be fixed, because the peaks were already cut off when it was recorded. Normalisation sets the loudness to a standard level, which matters if you are putting several recordings together, and the merge tool lets you do that without the join being audible.
Choosing, in one pass
- Keeping a master, or about to edit: WAV if the software is fussy, FLAC if it is not.
- Music to listen to, on anything: MP3 at 192 or 256, or AAC at 160 if you know it will play.
- Speech, to listen to: MP3 at 96 mono, AAC at 64 mono, Opus at 32 mono. Choose by what will play it.
- Sound out of a video, untouched: M4A, no re-encode.
- For a web page or a chat app: Opus, 64 to 96 kbps.
- Already lossy, and you want it in a different format: match the bitrate it already has. Do not go higher. Consider whether it needs converting at all.