An efficient and reliable algorithm for voiced/unvoiced discrimination is presented in this paper. Voicing determination is of great importance in many fields of speech processing, ranging from coding to recognition. In this work, the log value of the ratio between low and high frequency energies has been taken as basic parameter for the discrimination. Moreover differential parameters, computed as the difference of the log-energy ratios of adjacent frames, have been used. The algorithm is quite simple because it requires only two filters, two integrators and some logic for obtaining the final decision. The algorithm has been implemented in real-time on a TMS320C25 and it is shown that it requires only the 11% of the available time. Some classification examples are reported.
