Research on Segmentation Experience of Music Signal Improved Based on Maximization of Negative Entropy

With the rapid growth of digital music today, due to complexity itself, ambiguity definition category, and limited understanding characteristics human auditory perception, research on topics related automatic segmentation is still in its infancy, while infancy. Segmentation a prerequisite for fast effective retrieval resources, potential application needs are huge. Therefore, have important value. This paper studies an improved algorithm based negative entropy maximization well-posed speech separation. Aiming at problem that separation performance method depends selection initial matrix, Newton downhill used instead iteration as optimization find optimal matrix. By changing descending factor, objective function shows downward trend, dependence value reduced. The simulation experimental results show can separate source signal well under different values. average time reduced by 26.2%, number iterations 69.4%, both small. Fluctuations within range better solve sensitivity Experiments proved new significantly improve neural networks. Compared with existing methods, this excellent accompaniment singing separated music.