Research On Evaluation System Of Piano Sight-reading Ability Based On Music Theory

Sight-reading,which refers to the player playing directly according to the score,is an important part of piano learning.Due to the high professional requirements for teachers,traditional sight-reading learning is costly and inefficient.Using computers to evaluate sightreading music has significant advantages in reducing learning costs and improving learning efficiency.Recently,it has become a research hotspot,but there are still problems such as inaccurate evaluation results and non-conformity with music theory.This thesis aims to improve the music feature extraction method for sight-reading audio,and research a piano sight-reading ability evaluation system based on music theory standards.The system evaluates the user's sight-reading ability according to the scores of rhythm,main melody and musical performance of different levels of music score sight-reading audio to achieve the purpose of assisting piano instruction.The main innovations and research work of this thesis are as follows:(1)The method of sight-reading evaluation,according to the characteristics of music appreciation,we set two evaluation standards which are strength and weakness.The skeleton features such as rhythm and main melody are taken as the strong standard,requires that the sight-reading audio is the same as the original score,subjective feelings such as Musical emotion as weak standards.First,we define the Musical emotion as a point in the twodimensional Thayer plane in psychology,and train a two-dimensional prediction model of music perception based on support vector regression.Then,the distance between the Musical emotion of sight-reading audio and the Musical emotion of standard audio in the Thayer plane is used as the performance evaluation of sight-reading Musical emotion,to avoid ambiguity caused by using emotional words to evaluate the Musical emotion.Therefore,the accuracy and artistry of the evaluation results are improved.(2)On rhythm extraction,we use the interval of note fragments as the audio rhythm,and proposes a method for extracting note fragments using MFCC to correct errors.First,divide 88 frequency bands according to the main energy frequency of each note,detect the starting point of notes by multi-frame mean energy increment of each frequency band,and divide note segments with this,then use MFCC to design the note segment alignment algorithm,correct the detection error of note starting point.The detection accuracy of the algorithm reaches 98%,which improves the accuracy of rhythm evaluation.(3)On the main melody extraction,according to the characteristics of harmonic superposition and composition of piano notes,a main melody note extraction method based on "harmonic spectrum subtraction of left and right hand notes" is proposed.First,we determine candidate notes based on the music score,design a harmonic warping filter to obtain the harmonic spectrum of each note,and then select the left-hand note;Second,weighted minus the influence of the smoothed left-handed harmonics in the harmonic spectrum of the righthanded notes;Finally,the harmonic spectrum after spectrum reduction is analyzed to obtain the right-hand note,that is,the main melody note.This method makes full use of prior information and harmonic characteristics to reduce the error of note detection and algorithm F value is 92%.(4)We collected piano sight-reading scores,extracted their characteristics and built a system database,and implemented each scoring module to obtain a sight-reading ability rating of 0 to 5.Finally,we use the system client and online scoring platform to conduct experiments.The cosine similarity of the evaluation results given by this system and the manual is 0.99,which verifies the effectiveness and objectivity of the system's evaluation of the user's sightreading ability.