论文部分内容阅读
文中针对DCT字典和高斯随机观测矩阵对语音信号重构效果差的缺点,根据语音信号的特点,运用线性预测技术,构造适用于语音信号的LPC字典,接着通过Gram矩阵优化算法使用给定的LPC字典优化观测矩阵,提高重构性能,最后通过OMP算法对语音进行高质量重构。实验结果表明,采用线性预测字典与优化后的观测矩阵的重构效果比采用DCT字典与高斯随机观测矩阵的重构效果有大幅的提升。
Aiming at the shortcomings of DCT dictionary and Gaussian random observational matrix in reconstructing the speech signal, a LPC dictionary suitable for speech signal is constructed by using linear prediction technique according to the characteristics of speech signal. Then, using the LPC dictionary by Gram matrix optimization algorithm, Dictionary optimization observation matrix to improve the reconstruction performance, and finally through the OMP algorithm for high-quality speech reconstruction. Experimental results show that the reconstructed results of the linear predictive dictionary and the optimized observational matrix are greatly improved compared with the reconstructed results using the DCT dictionary and the Gaussian random observational matrix.