EMZETT.
Login

Audio formats: MP3, AAC, FLAC and Opus

In short: Audio formats differ in whether they compress losslessly or lossy and how good they sound at a small file size.

In more detail: Digital music arises through sampling: the signal is measured thousands of times per second (sample rate, for example 44.1 kHz on CD) with a certain bit depth (16 or 24 bits). Uncompressed (WAV), one minute of CD quality takes about 10 MB. Formats reduce this.

In Depth

Lossless

  • WAV, AIFF: uncompressed, huge.
  • FLAC (Free Lossless Audio Codec): compresses to about 50 to 60 %, without loss: unpacked, the original results exactly. Free and common for archives and hi-res streaming.
  • ALAC (Apple Lossless): like FLAC in the Apple world.

Lossy

They remove parts that human hearing hardly perceives (psychoacoustic model, for example masking of quiet tones by loud ones).

  • MP3 (1993): most widespread, works everywhere. At 128 kbit/s audible artefacts, 192 to 320 kbit/s good.
  • AAC: successor to MP3, better at the same bitrate. Standard with Apple, YouTube and streaming.
  • Ogg Vorbis: free, roughly at the level of AAC.
  • Opus: modern, free, very good at low bitrates and short latency. Used in video calls (WhatsApp, Discord, WebRTC) and streaming.

Bitrate and quality

The bitrate (kbit/s) determines size and quality. A lossy format at 256 kbit/s is, for most people and headphones, barely distinguishable from the original. Variable bitrate (VBR) adapts the rate to the material, constant (CBR) stays the same. Sample rates that are too high (96 or 192 kHz) bring no advantage for everyday listening; they make sense in production.

In practice

For archive and editing keep FLAC or WAV, for phone and on the go AAC/MP3/Opus. After each further lossy encoding quality falls; converting an MP3 to FLAC makes nothing better. Tags (title, artist, cover) are in ID3 or Vorbis comments. Tools: ffmpeg, Audacity, foobar2000. Licence: MP3 patents expired (2017), so it is freely usable.

See also: Video codecs, Compression, Microphone, Speech recognition and synthesis