Audio Codecs Explained: Music, Speech and Lossless Formats

A codec built for music and a codec built for a human voice are solving different problems, and knowing which one you are dealing with explains most of the odd things audio files do.

A dense music waveform and a sparse speech waveform side by side

Music codecs try to keep everything a good pair of headphones can resolve.

Speech codecs throw away almost all of that on purpose, because a voice only occupies a narrow slice of it.

That single difference is why a modern song sounds fine at 128 kbps, why a phone call sounds like a phone call, and why some old video clips play with a perfectly good picture and total silence.

Looking for the basics. If you are still working out what a codec is or the difference between codecs and containers, those two guides cover it properly. This one is about audio specifically.

Two Families, Built for Different Jobs

Audio codecs split into two families that are rarely explained side by side, even though the split explains a great deal.

Music codecs assume the sound could be anything - a cymbal, a cello, a voice, a car door. They have to cover the full range of human hearing, so they keep a wide frequency range and spend bits generously.

Speech codecs assume the sound is a person talking. Human speech lives in a narrow band and is produced by a vocal tract that works in fairly predictable ways, so a speech codec can model that machinery and transmit far less.

The payoff is dramatic. A speech codec can carry an intelligible voice at a fraction of what a music codec needs, which is why phone networks, radios and video calls have used them for decades.

The cost is equally dramatic. Play music through a speech codec and it comes out thin, warbling and unpleasant, because everything the model did not expect gets discarded.

The Music Codecs You Will Actually Meet

Four of them cover almost everything, and the useful ranking is not by quality but by what will open the file.

MP3, AAC, Vorbis and Opus compared on how widely they are supported

MP3 is the oldest and technically the weakest, and still the right answer whenever compatibility matters more than fidelity - see the MP3 encoder settings guide for the numbers worth using.
AAC sounds better than MP3 at the same size and sits inside most MP4 files, so you have been listening to it for years without choosing it.
Ogg Vorbis beats MP3 at the same bitrate and is open and patent free, which is why it turns up in games and open-source software.
Opus is the best of the four by a clear margin and the least likely to open in a random program.

Opus, Where the Two Families Merged

Opus is worth its own section because it is the one codec that refused to choose a side.

It contains both a speech model and a music transform, and it switches between them, or blends them, depending on what it hears. That is why it works at 6 kbps for a voice and at 128 kbps for an orchestra without changing codecs.

Every video call you have taken in the last few years almost certainly ran on it. The catch is the one on its own download page - it wins on quality per kilobit and loses on what will actually play it.

Lossless Audio, and Which One to Pick

Lossless codecs throw nothing away, so the decoded file is bit-for-bit identical to the source, at roughly half to two-thirds the size of the original.

Worth it for archiving a CD collection or for anything you will re-encode later, since repeated lossy encoding compounds the damage. Not worth it for a phone full of pop music.

Format	What it is	Use this when
FLAC	The open, standard choice, supported almost everywhere lossless is supported at all	Always, unless you have a specific reason not to
Monkey's Audio	Compresses tighter than FLAC and decodes more slowly, files ending .ape	You are archiving to disk and never streaming it
WavPack	Hybrid format that can store a small lossy file plus a correction file	You want one library that works on a phone and at home
TTA	Simple and very fast to encode and decode, files ending .tta	Real-time recording, or a low-powered machine
OptimFROG	The tightest compression of the group, and by far the slowest	Storage matters more than your time

FLAC is the default answer for almost everyone, and FLAC explained covers ripping, playback and tagging in depth.

Watch the .m4a extension. It is a container, not a codec, and it holds lossy AAC far more often than it holds Apple Lossless. You cannot tell which one you have from the filename, so check the file properties or open it in MediaInfo.

When an Old Clip Plays With No Sound

This is where speech codecs stop being trivia and start being your afternoon.

Old AVI files, camcorder clips, screen recordings and IP-camera footage frequently carry a speech codec that Windows has never shipped. The video decodes normally, the audio track is simply ignored, and you get a picture with silence.

Three reasons an old clip plays with no sound: audio codec tag 75, audio codec tag 45, and an unregistered ACM codec

Some players name the missing codec in the error, and the number is the thing to search for:

Audio Codec Tag 75. That is VoxWare MetaSound, an ultra-low-bitrate speech codec from the early 2000s - install VoxWare MetaSound and the track comes back.
Audio Codec Tag 45. That is G.726, common in older MPEG4 clips and IP cameras - the Sharp G.726 Audio Codec decodes it.
Acelp.net files. A VoiceAge speech codec that Microsoft once recommended for Windows Media Player, still fixed by the Acelp.net decoder.
Files ending .spx. That is Speex, the open speech codec Opus eventually replaced.

These are ACM codecs, which is an older Windows mechanism than the DirectShow filters most guides talk about. If one installs and nothing changes, registering a codec DLL or AX file is usually the missing step.

The One-Download Answer

If you would rather not identify the format first, one decoder covers most of the ground.

MPC Audio Decoder handles over thirty audio formats including AAC, FLAC, DTS-HD and Opus, which takes care of nearly everything a modern file will throw at you.

Pair it with MPC Audio Splitter if your lossless files will not open at all, since that is the component that reads FLAC, APE, TTA and WavPack containers before any decoding starts.

Neither helps with the legacy speech codecs above. Those predate this whole approach and still need their own small decoder.

Who Should Look Elsewhere

The tell is whether the file has a picture.

If the video will not play either, the problem is upstream of the audio and codecs and containers is the better starting point.

And if you only want to turn one audio format into another rather than understand any of this, the online audio converter does it in the browser, or convert between them with foobar2000 for a whole folder at once.

Quick questions

What is the difference between an audio codec and a speech codec?

An audio codec is built to carry any sound across the full range of hearing. A speech codec models the human vocal tract instead, which lets it use far less data but makes it unsuitable for music.

Which audio format should I use?

MP3 if other people or old devices need to open it, AAC for general use, and FLAC if you are archiving and want nothing thrown away.

Why does my old video have a picture but no sound?

Almost always a legacy speech codec Windows never shipped, common in old AVI files and camera clips. Check whether your player names an audio codec tag in the error, since that number identifies exactly which decoder you need.

Is lossless audio actually worth it?

For archiving a collection or for anything you will re-encode later, yes, because repeated lossy encoding compounds the damage. For everyday listening on ordinary equipment, most people cannot hear the difference.

Is .m4a lossy or lossless?

Either. It is a container that usually holds lossy AAC but can hold Apple Lossless, so the extension does not tell you and you need to check the file itself.

Why does Opus sound better but open in fewer programs?

It is much newer than MP3 and AAC, so a great deal of existing software and hardware was built before it existed. Quality per kilobit is where it wins; support is where it does not.