Objective Evaluation Of Singing Of Artistic Voice Based On Neural Network

Artistic voice is the life of people who take up the art job, estimation is so very important at the period of choosing and training person with acting ability that how to appropriate appraisal is pivotal.The way with scatheless and digital visualize and scientific that is used to analyse and appraise the artistic voice using computer and principle of acoustic is effective for appraising pronunciation.The singing voices were recorded from 48 young music students of singing who are from Guang'Xi normal university.The characteristics of singing,such as F1,F3,F0,vocal range,jitter, disturbance of F1, disturbance of F3 and average energy were assessed by way of voice analysis.BP Neural network analysis and wavelet Neural network were used to evaluate the singing voices objectively.The results were then compared with the subjective evaluations by experienced professionals in order to find better way for evaluate singing of artistic voice and the change law of every acoustic parameter that affect the evaluation quantitatively.Results:There were laws for the first singer that F1 was next to 450HZ,F3 was next to 2500HZ,vocal range was up to 3.5;F3 was among 2400HZ and 3200HZ; There were laws for the fifteen singer that F1 was next to 450HZ,F3 was next to 2400HZ; the first singer can be do well with every melody,the others can grasp the melody listed as F,E,G; the number of neural unit of wavelet Neural network for hidden layer was 2,the number of neural unit for hidden layer of BP Neural network was 27; wavelet Neural network analysis and BP Neural network were both in accord with the subjective evaluations by experienced professionals,furthermore the average errors of the former about the melody E and G were 2.51% and 3.86%,these of the latter were 2.59% and 4.14%;the weights that F3,F1,vocal range,F0,energy,perturbation of F3, perturbation of F1,jitter affected the evaluated result respectively were 60.58,55.764,45.434,42.249,33.642, 14.215,10.372,6.6258;vocal range positively affected evaluation ,F1 and F3 negatively affected evaluation.Conclusion: The acoustic parameters of the first singer show regular change; Neural network analysis can be used as an objective instrument to evaluate singing of artistic voices, the structure of wavelet Neural network was more simple than that of BP Neural network,and the former has higher correct rate; the order that the acoustic parameters affected the evaluated result listed as F3,F1,vocal range,F0,energy, perturbation of F3, perturbation of F1,jitter; This can be helpful to instruct,select and train professional singers.