Coding of telephone band speech at 5 kbit/s based on glottal excitation analysis

Application of a separated speech production model in low-bit rate speech coding is studied. The model consists of three parts: the glottal excitation, the vocal tract, and the lip radiation effect. These processes are first estimated with the AIF (adaptive inverse filtering) method. Coding of the glottal excitation is computed as a linear combination of two source signals: an impulse response of a second-order all-pole filter and a noise sequence. Combination of these two source signals is determined using the zero crossings rate of the glottal excitation obtained by the AIF method. The vocal tract transfer function is coded as an LPC (linear predictive coding) filter and the lip radiation effect is modeled with a fixed differentiator. Preliminary results that were obtained when this novel coding method was applied at 5 kbit/s are promising