Optimization of 2.4KBPS MELPe based on ARM9

MELPe (Mixed-Excitation Linear Predictive Enhancement) is an enhanced algorithm of MELP, which includes noise pre-processor and achieves better speech quality than MELP. To meet the demand of real-time embedded application, the MELPe algorithm is optimized based on ARM926EJ-S. The detailed process of optimization is presented, including three levels: algorithm optimization, code optimization and embedded assembly optimization. After optimization, the computation complexity is decreased by 87.6% (from 516.8MIPS to 64.1MIPS). And the PESQ test shows that the speech quality doesn´t descend. The experiments indicate that the effectiveness of the proposed approaches, which meets the requirement of real-time implementation.