An Automatic Scoring Method for Vocal Singing Based on Approximate Matching Algorithm

In this paper, the performance of approximate matching algorithm in large-scale music hum retrieval system is studied. In order to support the comparative study of several approximate matching algorithms, an algorithm independent test platform is developed. The basic algorithms studied include suffix tree, Hidden Markov Model (HMM), edit distance, Dynamic Time Warping (DTW) and unilateral continuous matching (OSCM). The music database used for the test included several datasets. By constructing 1500 different types of error query methods, a large number of experiments are carried out on the three algorithms, and their performance is analyzed and evaluated. The experimental results show that the algorithm based on unilateral continuous matching not only has a fast query speed, but also has a query accuracy of 100% when the user hum only contains the same direction as the melody contour error. When a user’s buzz contained two or fewer errors that were contrary to the melody’s outline, the median rate for the top 10 was about 90%. It is a query algorithm suitable for large cellular retrieval systems. Characteristic parameters such as pitch intensity, pitch and respiration of the test vocal music and the corresponding original sound were extracted, and each parameter was compared by the feature matching method, and an objective score was given according to the similarity degree through the scoring mechanism. The establishment of objective evaluation system of multimedia vocal music improves the intuitiveness and interactivity of Chinese vocal music teaching. The simulation results show that the algorithm complexity is low and the score is consistent with the subjective feeling.