Excitation modeling in a homomorphic vocoder

An efficient dynamic excitation modeling which emphasizes either the pitch-dependent periodic nature or the more random nature of the sound sources depending on the voicing state of the speech signal is applied to the homorphic vocoder. In this excitation modeling scheme, the cepstral peak value is used as the basis for classifying a given speech segment into three different classes: voiced, unvoiced, and mixed. Different excitation modeling strategies are used for the different classes. The algorithm is simulated, and clear improvement in the speech quality is shown