Implementation and evaluation of a 2400 bit/s mixed excitation LPC vocoder

The authors present an LPC (linear predictive coding) vocoder model with four features: mixed pulse and noise excitation, periodic or aperiodic pulses, pulse dispersion filter, and adaptive spectral enhancement. A 2400-bit/s speech coder based on this model has been implemented, as well as a 4800-bit/s coder which also includes Fourier series spectral information. Formal subjective test results show that this 2400-bit/s vocoder performs better than the government standard at the same bit rate, for both clean and noisy input speech. The quality of this coder approaches that of the higher-rate CELP (code-excited linear prediction) standard. The addition of Fourier magnitudes to the 4800-bit/s coder provides significant improvement, but both the bit rate and the computational requirements of the 4800-bit/s coder are much higher than for the 2400-bit/s vocoder. Therefore, this 4800-bit/s coder may not be as useful, but its performance does serve as a target for possible improvements to the parametric model.<>