Byte Block Audio


Byte Block Audio is a simple audio compression scheme for use with embedded devices. Multiple channels are supported, as are multiple bit-depth and sample-rates. There is a reference encoder for Windows, with source code included.


You can grab the latest version of BBA here. (July 15, 2008: see the change log at the bottom of this page.)


Public Domain



Byte Block Audio is a very simple & scalable type of ADPCM. BBA compresses a fixed number of samples into a single byte, hence the name [8^). Each sample is coded as a delta from the previous sample.

Each byte in the compressed bitstream has 2 parts, the "Adaptive" portion, and the "Delta Pulse Code Modulation" portion. Each delta is encoded using a fixed number of bits, then scaled with a "shift left" operation (SHL, to avoid multiplication). The SHL variable is called the "scaler" in the reference code. The "Adaptive" portion of the compressed byte will change the scaler, then the remaining bits of the compressed byte will be decoded to represent the delta from the previous sample. The initial sample value is set to 0, and the initial scaler value is set to 1.

The desired number of samples will control the distribution of bits int the compressed byte. See the following table:
Samples / Byte Adaptive Bits Delta Bits
So, for example using 3 samples per byte, the 2 most significant bits will adjust scaler, then the next two bits will define the delta for the 1st sample, then the following 2 bits will define the delta for the second sample, and the last 2 bits define the delta for the final sample.

There are two schemes for updating the scaler. In the 1-bit scheme scaler += 1 if the adaprive bit was 1, otherwise scaler -= 1. In the 2 bit scheme, scaler += adapt - 1. This lets the scaler decrease by 1, stay the same, or increase by 1 or 2. This distribution was chosen because it was deemed better to be able to overshoot the value, than to slowly ramp up to it.

The decoder is very simple, especially if you only want to decode a fixed compression level. One thing to notice is the output value is set to the average of the new value and the last value. This smooths out the output waveform, and is especially useful for the high compression levels where each sample only gets a 1-bit delta.

See the C++ source for any questions.


Change Log:

back to