Structure analysis of Chinese Peking Opera

Automatic processing of Chinese Peking Opera becomes more important than ever before. This paper presents a multistage system for structure analysis of Chinese Peking Opera. Taking advantage of opera theory, all procedures fits human's usual practice appropriately. Especially, a clustering technique is employed for collecting all concrete timbre types. The bottom-top strategy is also presented to reconstruct the structure which discriminates components of Peking Opera. Firstly timbre information sequence is obtained using Viterbi algorithm. There will be coherences or abrupt changes in this sequence generally. Then we gather component information of opera via Forward algorithm in a series of well-trained HMM. Finally, another Viterbi algorithm revised by greedy theory is applied to eliminate petty slices that smoothes the sequence and gets rid of possible errors. The proposed system has been tested on various Peking Opera recordings. These results are shown to demonstrate the performance of this system.