Evaluation Of Lute's Acoustic Quality And Recognition Of Lute's Musical Note Based On Artificial Neural Network

The research of evaluating musical instruments acoustic quality has a great contribution for the inheritance,development and popularization of Chinese national musical instruments,and it also produces many valuable advices for the improvement of musical instrument performance.Meanwhile,musical note recognition also plays an important role in instrument tuning,computer automatic transcription,music database retrieval,and electronic music synthesis.For the evaluation of musical instruments acoustic quality,the traditional method is based on the subjective evaluation of musical experts.However,the traditional method can hardly reflect the objectiveness musical instruments acoustic quality since the individual characteristics and environment factors result in great differences in evaluation results.For the recognition of musical notes,the traditional method is to match the estimated pitch with the standard frequency of musical notes.However,the following issues,such as low accuracy of notes recognition,poor robustness of recognition process,and narrow range of identifiable pitch,lead to a great challenging for musical notes recognition.In a case of lute,research of acoustic quality evaluation and musical note recognition based on artificial neural network is conducted in this paper.According to the subjective evaluation of musical experts,a lute acoustic quality evaluation optimization model was developed.In the proposed method,a music library consisted of 144 experimental samples of musical signals was built and the characteristic parameters(CC,CQT and MFCC)of musical tone signals were extracted and combined to six combined-features,and each combined-feature was input into neural network for network training,and the results of subjective evaluation were the target tag for monitoring online learning.Similar with the acoustic quality evaluation method,a musical notes recognition optimization model was developed based on the standard sound of musical notes.In the proposed method,a 25-note library consisted of 3600 lute notes signals was built,and the characteristic parameters(CQT and MFCC)of musical notes were extracted and combined to three combined-features,and each combined-feature was input into Softmax regression BP multi-classifier for network training,and the classfication codes of standard notes were the target tag for monitoring online learning.The experimental results show that the combined-feature(MFCC+CQT+CC)is the best representation of the lute acoustic quality,where the results of proposed method keeps high consistent with the subjective evaluation.The evaluation method used in the paper is very novel and feasible.Meanwhile,the combination features(MFCC+CQT)is the best representation of musical notes,which achieves high-precision note recognition of 25 kinds of musical notes from bass to treble,and the average recognition accuracy is 95.6%.Compared with other recognition algorithms,the recognition algorithm used in the paper has the advantages of less restricted conditions,more types of recognized musical notes and high recognition rate.