AAC vs Opus vs MP3

How lossy audio codecs work

All three are perceptual coders. They turn short windows of audio into the frequency domain. Then they use a model of human hearing to decide which parts can be dropped or coded roughly, because louder nearby sounds would mask them. The codecs differ in the transform, the window sizes, the hearing model and the entropy coding. Those differences decide how good each one sounds at a given bitrate and how much delay it adds.

Bitrate for audio is small next to video. A 160 kbps AAC track is a rounding error beside a 6 Mbps 1080p stream, so it is tempting to ignore it. The reason not to is compatibility. A video stream with an unsupported audio codec plays silently or not at all, as the container vs codec explainer shows.

The three codecs compared

Bitrate guidance reflects common practice and the codec developers' own recommendations rather than formal measurements.

AAC: the default for MP4 and HLS

Advanced Audio Coding was standardised in 1997 as the successor to MP3 and became the audio side of the MPEG-4 family. Its Low Complexity profile (AAC-LC) is what nearly every MP4 uses. HE-AAC adds spectral band replication for better results below about 64 kbps. HE-AAC v2 adds parametric stereo for very low rates. Every browser, phone and television decodes AAC-LC. Apple's HLS authoring guidance names AAC as the expected audio format, which is why streaming pipelines default to it.

For a stereo track in a video, 128 kbps AAC-LC is transparent for most speech. 160 to 192 kbps is a safe target for music. FFmpeg's built-in encoder is adequate. The Fraunhofer libfdk_aac encoder, where available, is slightly better at low bitrates. Whatever you choose, VideoBB's transcoder outputs AAC in its HLS renditions. That is what plays everywhere the HLS segments are sent.

Opus: the efficient modern option

Opus was published by the IETF in 2012. It merged Skype's SILK speech codec with Xiph's CELT music codec into one format. It switches between them, or blends them, depending on the content and bitrate. It is unusually good at low bitrates: clear speech at 24 kbps and convincing stereo music around 96 kbps. It also has very low latency, adjustable down to 5 milliseconds. That is why WebRTC requires it and why it is the codec inside almost every browser video call.

Opus is native to WebM and Ogg, and Matroska has carried it since its introduction. Its use in MP4 was standardised later and works in modern FFmpeg and in Chrome and Firefox. But older players and some Apple platforms are inconsistent about Opus in MP4 and in HLS. The safe rule: Opus with VP9 or AV1 in WebM, AAC with H.264 or H.265 in MP4. Opus always runs internally at 48 kHz, resampling other rates on input, which is invisible in practice.

MP3: everywhere, but past its best

MPEG-1 Audio Layer III dates from 1993, and its last relevant patents expired in 2017. It is decoded by every device ever made that plays audio, which keeps it useful for compatibility. At its 320 kbps ceiling it sounds fine. Below about 128 kbps its artefacts become obvious, as a swirling quality on high frequencies. AAC and Opus avoid that at half the rate. It also has a long algorithmic delay, no support for more than two channels in the original standard, and a hard 320 kbps limit.

MP3 is permitted in MP4 and in MPEG-TS, and old AVI files often hold it, so a remux can carry it through unchanged. For a new encode there is no reason to choose it when AAC is available in the same container.

MP3 still earns its place in one job: a stand-alone audio file for an unknown player. If you need just the soundtrack of a video, for a podcast feed or a transcript, the free extract audio tool saves it as MP3, M4A, WAV or Opus.

AAC or Opus for new encodes?

The choice usually follows the container and the destination.

Practical settings and traps

• Match sample rates sensibly. 48 kHz is the video standard and what Opus uses inside. 44.1 kHz is fine for AAC and MP3, but resampling back and forth is pointless.
• Keep channel layouts simple. Stereo is universal. 5.1 AAC plays on many devices, but browsers usually downmix it. Opus supports multichannel, but few video players expose it.
• Watch for AC-3 and DTS in files from discs or capture cards. Both are legal in MP4 and MKV, and both fail silently in most browsers. Transcode them to AAC while copying the video: ffmpeg -i in.mkv -c:v copy -c:a aac -b:a 192k out.mp4.
• Audio bitrate is a tiny fraction of the total, so saving on it is a false economy. The bitrate guide shows where the real budget goes.
• If lip sync drifts after conversion, the cause is nearly always variable frame rate video, not the audio codec. The AVI conversion guide covers the fix.
• If a clip's sound cannot be saved, or you have no rights to the music, remove the track instead. The free mute video tool drops the audio and copies the video without re-encoding it.