TY - GEN
T1 - Regional recognition of Chinese folk songs based on LSD audio segmentation algorithm
AU - Li, Juan
AU - Wang, Yinrui
AU - Yang, Xinyu
N1 - Publisher Copyright:
© 2017 ACM.
PY - 2017/2/18
Y1 - 2017/2/18
N2 - This paper discusses a method of studying the regional recognition of Chinese folk songs. This is not only of great significance to music recommendation based on regional styles, but it is also conducive to discovering music characteristics and regional styles of specific geographical folk songs. First, fully considering the integrity of musical expression when extracted features, the Layered Stabilities Detection (LSD) audio segmentation algorithm we have proposed is used. Then, HMM-GMM regional recognition models are built for each region's folk songs by considering the dynamic time information in them. Finally, new folk songs are recognized by measuring the probability of each model and choosing the model with the largest probability as their region. Experiments show that the accuracy of our regional recognition algorithm is 83.14%, an improvement of at least 7.81% compared with the SVM, GMM, and K-NN regional recognition models.
AB - This paper discusses a method of studying the regional recognition of Chinese folk songs. This is not only of great significance to music recommendation based on regional styles, but it is also conducive to discovering music characteristics and regional styles of specific geographical folk songs. First, fully considering the integrity of musical expression when extracted features, the Layered Stabilities Detection (LSD) audio segmentation algorithm we have proposed is used. Then, HMM-GMM regional recognition models are built for each region's folk songs by considering the dynamic time information in them. Finally, new folk songs are recognized by measuring the probability of each model and choosing the model with the largest probability as their region. Experiments show that the accuracy of our regional recognition algorithm is 83.14%, an improvement of at least 7.81% compared with the SVM, GMM, and K-NN regional recognition models.
KW - Audio segmentation
KW - HMMGMM
KW - Regional recognition
KW - Stabilities
UR - https://www.scopus.com/pages/publications/85020844042
U2 - 10.1145/3057039.3057049
DO - 10.1145/3057039.3057049
M3 - 会议稿件
AN - SCOPUS:85020844042
T3 - ACM International Conference Proceeding Series
SP - 60
EP - 65
BT - Proceedings of 2017 9th International Conference on Computer and Automation Engineering, ICCAE 2017
PB - Association for Computing Machinery
T2 - 9th International Conference on Computer and Automation Engineering, ICCAE 2017
Y2 - 18 February 2017 through 21 February 2017
ER -