Concurrently, research in data compression has explored dynamic and forward-looking approaches that build on traditional methods such as Huffman coding.

Huffman coding would be the easiest thing to implement this way, but it also won't exploit a lot of your compressibility. Just build up a frequency table from characteristic data and use that to ...

While many bit-packing algorithms exist, the compressor here uses a technique called block floating-point (BFP) encoding. BFP generates one exponent per N redundancy-remover output samples.