Decomposition of speech signals into a deterministic and a stochastic part

A novel method enabling efficient decomposition of the speech signal into a deterministic and a stochastic part is presented. The deterministic part is modeled by harmonically related sinusoids; the proposed method makes use of a third order polynomial with real coefficients for the harmonic amplitudes and the phase is assumed to be linear. To obtain the stochastic part the deterministic part is simply subtracting from the original speech signal in the time domain. The results obtained on a large speech database demonstrate effective decomposition of the speech signal between the two parts