AN ITERATIVE ALGORITHM FOR DECOMPOSITION OF SPEECH SIGNALS INTO PERIODIC AND APERIODIC COMPONENTS

The speech signal may be considered as the output of a time-varying vo cal tract system excited with quasiperiodic and/or random sequences of pulses, The quasiperiodic part may be considered as the deterministic or periodic component and the random part as the stochastic or aperio dic component of the excitation, In this paper, we discuss issues invo lved in identifying and separating the periodic and aperiodic componen ts of the source. The decomposition is performed on an approximation t o the excitation signal, instead of decomposing the speech signal dire ctly, The linear prediction residual signal is used as an approximatio n to the excitation signal of the vocal tract system, Speech is first analyzed to determine the voiced and unvoiced parts of the signal, Dec omposition of the voiced part into periodic and aperiodic components i s then accomplished by first identifying the frequency regions of harm onic and noise components in the spectral domain, The signal correspon ding to the noise regions is used as a first approximation to the aper iodic component, An iterative algorithm is proposed which reconstructs the aperiodic component in the harmonic regions. The periodic compone nt is obtained by subtracting the reconstructed aperiodic component si gnal from the residual signal, The individual components of the residu al are then used to excite the derived all-pole model of the vocal tra ct system to obtain the corresponding components of the speech signal, Experiments were conducted using synthetic speech, They demonstrated the ability of the algorithm for decomposition of a synthetic speech s ignal made of a mixture of periodic and aperiodic components, Applicat ion to natural speech is also discussed.