Singing Grading Method In Light Of Feature Parameters

As the rapid development of economy and people's cultural life, many singing contest programs come out. However, most of scoring methods are based on a number of areas of expertise, professionalism, and personal preferences, so that the result of scoring will inevitably mixed subjective factors and rise to a lot of controversy. On the other hand, some existing amusing singing score softwares contain the problem of irrational parameters, so that the scoring result are irrational and limited. In view of the above problem, we propose a more comprehensive and rational scoring method.Through the research and analysis of voice material, with the communication and discussion with some professionals on music and singing, we choose accent, lyrics and intensity of sound as the indicators to reflect the proficiency of a singer. These feature parameters such as pitch, MFCC and the average amplitude of sound signals are considered as the correspondences to those indicators above. Pitch frequency can be extracted by auto correlation function (ACF). To get MFCC, it should make use of FFT, Mel-filter, DCT. By calculating the average of all the sampling points of the amplitude of the absolute value, the average amplitude of sound signals can be obtained.The scoring idea is based on the comparison of the feature parameters, the voice of singer as the target sample signal file and the voice of original singer as the standard sample signal file. We compare the difference of the feature parameters between target sample and standard sample which can be obtained by using dynamic time warping (DTW) to compare the two samples. The less the difference is, the high the score of a singer will be, and vice versa. Due to the singing difficulty is reflected by the intensity of sound, we suggest the score be proportional to the intensity of sound. According to comparison results and the intensity of the target sample, we propose a formula to reflect the proficiency of a singer by integrating the factor assessment method.Through the simulation with MATLAB, the results illustrate that this scoring method can reflect the level of singer in accent, lyrics and other aspects better than other singing scoring softwares mentioned in this Thesis, and it is consistent with the subjective feeling of people.