Research Of Key Technologies Of Music Intelligent Sparring Partner

Music practice apps are an important direction in the field of music education.However,there are some limitations to music practice,such as the wear and tear and absence of sheet music,time and location restrictions,and differences in the quality of practice teachers.To address these issues,optical music recognition can convert traditional paper sheet music into electronic sheet music,making it easier for performers to learn and practice;music performance assessment can analyze and evaluate music performances,identify performers’ errors and weaknesses,and help them progress faster.Therefore,this article conducts research on optical music recognition and music performance assessment,solving or alleviating existing problems based on existing work.The specific research work of this article is as follows:Firstly,this article proposes a lightweight optical music recognition method CRNN-lite,which achieves the dual goals of lightweightness and accuracy.Specifically,CRNN-lite introduces residual depth separable convolutions into the convolutional layers to reduce computational complexity and accelerate feature map extraction.It also uses simple recurrent units in the recurrent layers to avoid the strong dependency problem of serial computing.In the transcription layer,the parameters of the cross-entropy function are adjusted to specialize learning imbalanced sample data.Experimental results show that this method can effectively improve training speed,with a single iteration taking43% as long as the baseline network.The symbol error rate on distorted image data is 1.12%,and the sequence error rate is 14.5%,significantly above the comparison scheme.Secondly,this article proposes a music performance assessment method based on signal processing,which refines the evaluation granularity and assigns scores to musical measures.Specifically,the method uses the p Yin algorithm to extract the fundamental frequency of the original audio and playing audio,further converting them into MIDI sequences.It uses the section-sequence dynamic time warping(SSDTW)algorithm based on the playing measure to determine the sequence overlapping interval between the playing music and the standard music,and finally assesses the music melody and rhythm in that interval.Experimental results show that the accuracy of pitch recognition on the dataset of the p Yin algorithm is better than that of the comparison method.At the same time,under the effect of time factors,the sequence matching algorithm based on SSDTW can improve the matching coincidence rate of musical measures to above 98.6% in cases such as variations in volume,mistakes,and changes in key.In cases of variations in rehearsal,the matching of musical measures also improves to varying degrees.Finally,the proportion of effective musical measures can reach as high as 82.4%.