One Codec for Every Voice
Every voice note, every video call and nearly every stream in your browser passes through one codec. It listens for what you will not miss.
Put on headphones if you have them, and press play.
These are 12 seconds of sound as a studio keeps it, with 48,000 measurements a second for each ear and each one a 16-bit number. It comes to 2.30 MB.
B is the same recording in 43.5 kB, which is 53 times smaller. Switch between A and B while it plays.
Most people cannot tell them apart. Further on you can find out whether you can.
How do you throw away most of a sound? The obvious way is to measure it less often.
Try a few settings. The high notes go first, and after them the consonants.
Or you can measure less carefully, and round each number to fewer bits.
Now a hiss rides on everything. The hiss is the rounding.
Do both at once and you are down to 64 kb/s, and it sounds dreadful.
Now listen to a real codec that was given exactly the same number of bits.
A codec does not throw numbers away evenly.
It works out which pitches you are listening to and spends its bits there. A quiet sound close in pitch to a loud one cannot be heard at all, so it gets almost nothing.
This is everything that was taken out of the recording you just heard, and nothing else.
It is not silence. It is what the codec judged you would not hear with the rest playing over it.
Here is the same sound as a landscape. Time runs from left to right, pitch runs away from you, and height is loudness.
Switch from A to B and watch the far ranges, where the highest pitches are, go flat.
By 2012 the internet used one codec for telephone calls and another for music, and paid patent fees on both. Opus replaced the pair with a single free format that keeps two ways of listening.
One treats sound as a voice, a throat and a mouth shaping a buzz. The other treats it as a mixture of pitches. The chart shows which it chose for this recording at each size, and how high a pitch it kept.
Now squeeze it yourself. Drag the dial while the sound plays, and stop where it begins to bother you.
The same size from three generations. MP3 dates from 1993, HE-AAC from 2003 and Opus from 2012.
Choose each in turn and listen to it as B. At this size the older two are plainly struggling.
Knowing which is which makes differences easy to imagine. A blind test takes the knowledge away.
X is secretly either A, the original, or B, Opus 32 kb/s. Listen to all three as often as you like, then say which one X is.
This version climbs. It begins with the smallest file and moves up each time you prove you can tell it from the original.
It stops at the first file you cannot tell apart. That is your own limit, on your own equipment.
At six thousand bits a second the file is 279 times smaller than the original, and there is no room left for music. Opus keeps the one thing that still fits, which is speech you can understand.
MP3 at its lowest setting sounds like a telephone under water.
For engineers, these are the other controls the encoder offers, changed one at a time while everything else is held still.
Choose a row to open that file in the viewer.
Sound is one number after another.
A picture is a great many numbers side by side, and the eye forgives different things from the ear.