| With the improvement of life,people pay more and more attention to the cultivation of musical accomplishment.Piano,the king of musical instruments,is favoured by many people.But beginners usually can’t realize the mistakes made in their playing.If there is a technology,which can identify the audio files generated by piano playing and compare them with standard electronic scores,it will surely have important practical value and wide application.As a branch of pattern recognition technology,piano music recognition has developed rapidly with the promotion of machine learning technology in recent years.This thesis combines traditional time-frequency domain audio analysis technology and convolutional neural network recognition technology to study the recognition and error correction technology of piano playing music.The contribution of this thesis is as follows:Firstly,the audio file is analyzed in the time-frequency domain,and an endpoint detection algorithm is designed to identify the start and end points of each note,obtain single notes,and draw their spectrogram.Aiming at the shortcomings of the traditional dual-threshold endpoint detection algorithm that relies too much on thresholds,an endpoint detection algorithm based on short-term energy difference is proposed.By finding the short-term energy difference peak to determine the starting point of the note,two layers of judgment are designed to determine each starting point The corresponding end point reduces the dependence on the threshold and improves the accuracy of the algorithm.Secondly,the spectrogram of each note is generated with time as the horizontal axis,frequency the vertical axis.The hue or gray value of the points in the figure indicates the strength of a given component at a given moment,which intuitively reflects the time of the audio and frequency domain information.Then a convolutional neural network is designed and a data set is constructed for training.The experiment proves that the accuracy rate of 93.5% can be achieved through the classification and recognition method based on the note spectrogram.Finally,based on the study and analysis of MIDI files,an algorithm is designed to read the standard information in the MIDI file and compares it with the time value and pitch information identified in the previous two steps to complete the error detection of playing music. |
