In this thesis we present a new algorithm of voiced/unvoiced/silent detection of speech. The decision algorithm is based on the results of zero crossing countings of linear delta modulation (LDM) bit stream and band-pass filtered output of decoded LDM signal. To optimize the system parameters and to test the validity of the proposed algorithm, computer simulation has been done extensively with various real speech of differing characteristics. The simulation results show that the new speech detection scheme works accurately. Based on the algorithm developed, a hardware systems has been designed for use in a digital speech interpolator. An efficient scheme of hardware realization without using a multiplier is suggested.