Digital Sound & Music: Concepts, Applications, & Science, Chapter 5, last updated 6/25/2013
126.96.36.199 Lossless Audio Compression
Lossless audio codecs include FLAC, Apple Lossless, MPEG-4 ALS, and Windows Media
Audio Lossless, among others. All of these codecs compress audio in a way that retains all the
original information. The audio, upon decompression, is identical to the original signal
(disregarding insignificant computation precision errors). Lossless compression generally
reduces an audio signal to about half its original size. Let’s look at FLAC as an example of how
Like a number of other lossless codecs, FLAC is based on linear predictive coding. The
main idea is as follows:
Consider the audio signal in the time domain.
Come up with a function that reasonably predicts what an audio sample is likely to be
based on the values of previous samples.
Using that formula on n successive samples, calculate what n+1 “should” be. This value
is called the approximation.
Subtract the approximation from the actual value. The result is called the residual (or
residue or error, depending on your source).
Save the residual rather than the actual value in the compressed signal. Since the residual
is likely to be less than the actual value, this saves space.
The decoder knows the formula that was used for approximating a sample based on
previous samples, so with only the residual, it is able to recover the original sample value.
The formula for approximating a sample could be as simple as the following:
where ̂( ) is the approximation for the
sample and is the predictor coefficient. The
residual ( ) is then
( ) ( ) ̂( )
The compression ratio is of course affected by the way in which the predictor coefficient is
Another source of compression in FLAC is inter-channel decorrelation. For stereo
input, the left and right channels can be converted to mid and side channels with the following
Values for mid and side are generally smaller than left and right, and left and right can be
retrieved from mid and side as follows: