Motorola advanced excitation model vocoder

In this paper, we introduce the Motorola Advanced eXcitation Model (MAXM) vocoder. This speech compression algorithm incorporates neural network control, robust parameter extraction methods, and time domain orthogonal error models that provide for improved speech quality and greater flexibility relative to prior art excitation-based coding methods. Furthermore, the vocoder structure provides for reduced computational complexity, identity system capability, and low sensitivity to acoustic interference. The MAXM error model configuration is well suited for adaptation to multiple data rates without changing underlying modeling methods.