为改善哼唱检索系统中利用旋律轮廓和节奏进行匹配的性能,提出一种新的联合音高与能量的音符切分算法。该算法改进基于自相关的基音提取算法,对提取的基音频率曲线进行后处理,并在切分过程中保持能量的分割信息,利用半音曲线的突变做切分,以提高音符切分的准确度。实验结果表明,在安静实验室环境下,该算法能获得88.75%的分割准确度。
In order to improve the retrieval performance of using melody contour and rhythm in Query By Humming(QBH), this paper proposes a new algorithm using pitch and energy for note segmentation. Autocorrelation pitch extraction algorithm is improved. Post processing is used to smooth the pitch frequency curves, and at the same time, energy segmentation information is maintained in the process of note segmentation. Semitone curve is used for segmentation. This algorithm improves the accuracy of the note segmentation. Experimental results indicate that the accuracy rate of the algorithm is 88.75% in quiet laboratory environment which is of great significance in QBH.