Research On The Extraction Method Of Music Melody And Its Application

Music melody is a sequence of fundamental frequency values corresponding to the pitch of the predominant source.As senior semantics of music,it can be used to the description of music content,and can be used as a front-end or intermediate processing steps for query-by-humming system,music genre classification and music emotion recognition,which makes the extraction and application of music melody become a research hot spot in the field of music information retrieval.This paper focuses on salience based music melody extraction methods and the application of musical melody in music emotion recognition.The main research contents are listed as follows:1.Aiming at the problem that the discontinuity of the pitch sequence of the same sound source is caused by the interference of different sound sources in polyphonic music which reduces the accuracy of pitch estimation,an improved pitch contour creation and selection algorithm for melody extraction is proposed.Firstly,a method based on auditory streaming cues and the continuity of pitch salience is proposed to create pitch contour by using the continuity of pitch salience.In order to further select the melody pitch contour,a dynamic time warping algorithm is proposed to calculate the similarity between the melody and non-melody pitch contours to remove non-melody pitch contours.Finally,the octave errors in the melodic pitch contours is detected based on the long term relationship of the adjacent pitch contours.Simulation experiments were performed on the datasets ORCHSET and MedleyDB respectively.The results show that the proposed algorithm improves the overall accuracy by 3.32% and 3.77% respectively.2.In order to improve pitch estimation accuracy,an improved melody extraction algorithm based on pitch salience is proposed.Aiming at the interference of the accompaniment or harmony of the instrument in music,the fundamental frequency of the predominant source is distorted or lost.According to the continuity of the pitch salience and the stability of the higher harmonics,a new method of creating pitch contours is proposed that based on pitch static likelihood function and pitch salience dynamic likelihood function.To take advantage of the timbral inconsistency of different sound sources in extracting melody pitch contour,the mel-frequency cepstral coefficients of the pitch contour is proposed as the timbre feature and the timbre characteristics are also calculated from the harmonic amplitudes of the pitch contour.The simulation experiment of the improved algorithm shows that the algorithm improves the pitch estimation accuracy and overall accuracy.3.The change of music emotion over time is not uniform,and in order to abstract the features of adjacent emotion change points,a dynamic music emotion recognition algorithm based on melody extraction and convolutional neural network was proposed.Firstly,the algorithm uses the melody extraction method to divide the music signal,and extracts the frame-level features of each segment.Then,the segmentation feature is abstracted by the structural model based on the convolutional neural network.Finally,the output of the network was concatenated with the melody contour feature of the corresponding segment,and input the regressor to get the emotional value of the emotional valence/arousal values based on theemotional model.Simulation experiments on data set DEAM15 show that the proposed method can effectively improve the recognition accuracy of music emotion compared with other recognition algorithms.