Simulation of the multi-band mixed excitation linear predictive speech coder at 2.4 kbit/s
Wang Du · Jounal of Xidian University · 2000
Based on the traditional LPC parameteric model, five additional features, mixed pulse and noise excitation, aperiodic pulses, adaptive spectral enhancement, pulse dispersion and Fourier magnitude modeling are introduced to further improve the quality of synthesized speech. The computer simulation and informal listening test results show that the new Multi Band Mixed Excitation Linear Predictive ( MBMELP ) speech coder at 2.4 kbit/s can produce a synthesized speech quality close to that by CELP algorithm at 4 8 kbit/s .