Gradient Encoding for Low-Bit-Rate Stored Speech Applications

For stored speech application, one of the ways of generating efficient binary data is tree encoding, which examines and verifies the sequences of a prespecified number of bits by varying the data in every possible combination and selecting the one that yields the best signal-to-noise ratio. This way of exhaustive searching for the best bit pattern demands a large number of computations, and the number of computations expands geometrically as the number of bits in the tree (i.e., the number of sequential bits chosen to explore the range of variation of the decoder voltage) increases. This leads to a further uncertainty about whether the number of bits chosen is satisfactory or not for any given section of speech.