A maximum entropy encoder for speech

A new approach to the encoding of speech signals is proposed based on directly maximizing the first order entropy of the data sent down the channel. It is formulated recursively and can be implemented via lookup tables. A corresponding decoder is also derived and can be implemented in a similar manner. The performance of the system is competitive with other methods but differs qualitatively as the design is based on entropy considerations rather than minimizing mean square error as is done is most other methods.