Research Of Score Following Based On DTW Algorithm Combined With Onset Detection

Audio analysis plays a significant role in multimedia information processing.Music is an art composed by organized and regular musical sounds, which expresseshuman feelings through singing songs or playing musical instruments. The content ofmusic is full of ideals and reflects the real world. People can’t live without music.With the high-speed development of computer science, people have higher demandsfor computers, hoping computers realize automatic music transcription during musicalperformance. Score following is an important part of automatic music transcription,which helps computer receive music on live and eventually locates the right positionof note in score. To solve the problems that Dynamic Time Warping (DTW)algorithm of score following has low time accuracy and can’t detect error notes, wedesigned an algorithm based on modified DTW.This paper combines the knowledge of music theory, signal processing andpattern recognitions to propose a system based on DTW algorithm and combined withonset detection algorithm, which extracts the chroma feature based on Constant-QTransform (CQT), and then compares the two feature vectors of audio signals. Finally,using the DTW algorithm realizes alignment between audio and score, whichdetermines playing notes of each moment. The main work in this paper includes thefollowing aspects:Firstly, framing in the system of score following based on time turns intomonophonic or polyphonic, in order that the algorithm is not affected by rhythm andhas better robustness. It lays a foundation for detecting error notes.Secondly, we studied the chroma feature in the field of score following andintroduced the specific algorithm. This paper uses the chroma feature based on CQT,which is more in line with the music theory characteristics of the scale frequency withexponential distribution.Finally, we compute the Euclidean Distance of two groups of feature vectors, andthen figure the optional alignment path using DTW arithmetic. According to theanalysis of path, we can determine playing notes of each moment. The results showthat the proposed algorithm compared with traditional method improve by7.64%on precision of alignment time and at the same time it can ignore the impact of beatvariation and detect error notes during the audio.