Article Preview
Top1. Introduction
Reforms in computer science have driven the development of online web-based teaching, which has led to the diversification of vocal music courses. A large number of institutions have introduced online vocal teaching courses, which can help students to correct their vocal style and improve their model singing ability, make it easier to scientifically fulfil the training objectives of music majors, help students to better understand music, and occur an essential position in the quality education of students. However, there are many problems that cannot be ignored in the process of training musicians and students learning vocal music in schools. Different students have different basic musical qualities and it is difficult to achieve the teaching objectives by relying on teachers to require uniform educational work. It is also difficult for teachers to analyses and understand each student comprehensively, which does not ensure that students receive targeted training in learning vocal music (Li 2021, Fu 2021). The basic idea of musical sound recognition is to extract the vocal characteristics of a musical instrument and input them to a classifier for musical sound recognition, the ability of which depends on the performance of the musical features and the discriminatory ability of the recogniser. With the advent of deep learning techniques, the development process of audio recognition technology has been significantly accelerated. Computer technology can save a great deal of energy, time and financial resources to replace the teacher in the task of identifying different musical notes in a vocal music teaching course, and computer algorithms can also perform the identification of singing and even automate the creation of musical scores. For the same instrument, the notes, loudness and timbre can bring different musical feelings. With the continuous development of music information retrieval technology, there is a rich variety of musical characteristics, and the existing music recognition technology is mostly a simple combination of all musical characteristics, but this does not meet the high technical requirements of music recognition. The Mel frequency cepstral coefficient and linear predictive cepstral coefficient are important technical tools for the study of the cepstral domain, and the selection, screening and combination of features are the next research focus of the current music recognition model. On the other hand, there is a large amount of noise interference in the music teaching curriculum, which leads to the performance of existing audio classifiers not meeting the practical needs. Most of the current research methods are still based on traditional signal processing methods, and there is less research on music recognition using deep neural networks, and there is still much room for improvement in terms of recognition accuracy and efficiency. In particular, the Mel frequency cepstrum coefficients are mostly selected manually, making the development of the basic feature recognition module of the music recognition model more restricted; and the traditional sound recognition algorithm model is still not good enough in terms of recognition accuracy and efficiency as well as the detection of abnormal noise. Against this background, in order to enhance the ability to extract musical tone features and the perception of timbre recognition, improve the accuracy of musical tone recognition, design a high-performance musical tone recognition model; and combine it with the vocal music teaching course to improve the teaching shortcomings of the traditional vocal music teaching course and realise the informatization of the vocal music teaching course, the research uses multi-layer perceptron and Meier's inverse spectral coefficient for musical tone feature recognition, changing the The fusion of multiple network models differs from a single network model in that it is more resistant to noise interference and has higher recognition accuracy. The article is divided into four parts: the first part summarises and outlines the relevant research and studies in this field and analyses the shortcomings of existing research; the second part provides a detailed description of the proposed methodology; the third part discusses and analyses the experimental results of the performance verification; and finally, the conclusions of the study are outlined.