论文部分内容阅读
在批处理音频信号盲分离过程中,要将分离出的语音信号帧拼接成连续的语音比较困难,这主要是由于盲信号分离存在分离信号排列顺序的不确定性。笔者根据语音信号短时平稳特性,利用基音周期为窗长,连续分割盲分离信号,将各个窗内数据的归一化自相关函数值取平均作为音频信号的模式特征,最后根据相似性阈值和最小距离原则进行信号聚类分析,从而克服提取盲分离语音信号中的信号顺序不确定性,获得连续语音信号。实验仿真证明了该方法的有效性。
In the process of blind separation of batch audio signals, it is difficult to spliced the separated speech signal frames into continuous speech. This is mainly due to the uncertainty of the separation order of blind signals. According to the short-time stationary characteristics of the speech signal, using the pitch period as the window length, the blind segmentation signal is divided continuously and the normalized autocorrelation function values of each window are averaged as the pattern features of the audio signal. Finally, according to the similarity threshold Minimum distance principle of signal clustering analysis, thereby overcoming the blind separation of speech signal in the signal sequence of uncertainty, access to continuous speech signal. Experimental simulation proves the effectiveness of the method.