An implementation of realtime speech speed control system using hierarchical cascaded integrator and comb filters

Abstract

In this paper, our algorithm for speech speed control system and its DSP implementation are described. The system slows or speeds up speech without changing its pitch, which aids learning a foreign language, listening for older people, and so on. The proposed algorithm is based on the phase vocoder, which doesn’t need extraction of voiced fragments from a speech signal. And the reconstruction of time-scaled signal involves no smoothing operation between voiced and unvoiced sections. The speech signal is decomposed to bandpassed frequency channels so that the envelope of each of the channels is stretched. Then the stretched envelope is multiplied by the corresponding sinusoidal, and by summing up the channel components the speed controlled signal is reconstructed. Each band has very narrow bandwidth, which requires a large amount of computation with a FIR-based design. To achieve realtime operation of our system, a hierarchical structured cascaded integrator and comb (HSCIC) filter is also introduced, which can have highly sharpened characteristics with only a few multipliers.