Recently, we have proposed a novel method to describe digital sound data by means of rule dynamics in two states and three neighbors cellular automata. In this paper, we report the results of evaluation about the reproducibility and the performance of the sound description by calcuating distortion rate, power spectra and return map of signal amplitude. A typical result in our numerical evaluations gives 21.29dB as averaged distortion rate. Comparison between the two return maps, the one taken from the original signals and the other from the amplitude data generated by the present method shows some difference, which indicates that the present method of rule extraction is not yet optimized in describing sound data. In order to improve the fidelity of describing the original data, we have tried to apply all the possible rule sequences which are generated by combining three rules within applying them eight times to a certain initial bit pattern. In the best case, the trial can give strong possbility of describing digital data wih high fidelity and with considerable compression.