论文部分内容阅读
为了提高高频区域时频单元标记和听觉分割的准确性,提出一种改进听觉组织的单声道浊语音分离算法.在组织阶段中,首先该算法利用不同的特征对高频和低频中的时频单元进行标记.增强包络自相关函数被用来标记高频区域的时频单元.然后,利用起始和截止分析得到听觉片段,起始和截止分析方法可以有效地将语音和噪声分割到不同的片段.根据已经分离的浊语音二值模将这些片段选择性地重组到目标流中.系统评估表明,该算法优于原来的系统.
In order to improve the accuracy of time-frequency unit labeling and auditory segmentation in high-frequency region, a mono-voiced speech separation algorithm is proposed to improve the auditory tissue. In the organization stage, the algorithm first uses different features to analyze the high frequency and low frequency Time-frequency units are marked.Enhanced envelope autocorrelation function is used to mark the time-frequency unit in the high frequency region.Then, the auditory segment is obtained by using the start-up and the cut-off analysis, and the start-and-stop analysis method can effectively segment the speech and noise To different fragments, and selectively reassembled these fragments into the target stream based on the voiced binarial model which has been separated. The systematic evaluation shows that the algorithm is superior to the original system.