摘要： 时频比是混合信号在时频域幅值特性的比值,利用时频比寻找混合信号中的单源点,对相应的比值构成的矩阵求逆可以得到对源信号的估计。针对基于时频比的盲源分离将信号变换到时频域后计算量大且对算法有效的时频窗较少的问题,提出用重复结构周期内的时频点代替整个时频域进行单源点的检测,重复结构内的时频点在每个周期内都有相似的值,通过减少一个周期内时频点的检测,由单源点对应的时频比恢复出源信号。用相似 系数矩阵评价分离效果,仿真实验结果表明,在达到几乎相同的相似系数的情况下,运行时间可减少45. 43% ,可有效降低运算量。

关键词: 时频比, 单源点, 线性时频变换, 重复结构, 节拍谱, 语音音乐信号分离

Abstract: Time-frequency ratio is the ratio of mixed signals’ amplitudes in time-frequency domain. The inversion of the matrix composing of the corresponding time-frequency ratios which is gotten from single source points in mixed signals can obtain the estimate of the source signals. Blind source separation based on time-frequency ratio has a large amount of calculation and small number of effective time-frequency windows after transforming the signals to the timefrequency domain. Time-frequency bins comprising the repeating patterns have similar values at each period. According to this characteristic,analysis zones including a repeating period are used to detect single source points instead of the whole time-frequency domain. Using similarity coefficient matrix as the separation effect standard,simulation experimental results show that the proposed method can reduce 45. 43% of time with the same separation accuracy.

Key words: time-frequency ratio, single source point, linear time-frequency transform, repeating structure, beat spectrum, speech-music signal separation

中图分类号: 

• TN912. 34