Wavelet decomposition is a widely used method to preprocess temporal signals before they could be analyzed by Artificial Spiking Neural Networks (ASNN). This study proposes a biological plausible way to encode the temporal signals into spike trains with wavelet amplitude spectrum represented by the delay phases during each encoding period. The encoding method is presented in the form of a spiking neuron model for easy implementation in ASNN. The proposed neuron model is tested on encoding of human voice records for speech recognition purpose, and compared with results from continuous wavelet transform. The nonlinearity properties and choices of biological plausible wavelet kernels for the proposed encoding method is discussed for the generality of its application.
