人工智能语义分割技术在钢琴教育中的应用研究

Application of Artificial Intelligence Semantic Segmentation Technology in Piano Education

目前,素质教育越来越被重视,作为素质教育代表的音乐教育也越来越被关注,但是音乐教育却极大受限于人工教育资源。人工智能在音乐教育中的辅助,从计算机的角度讲就是信号类型转换的过程。例如对于学者弹琴,需要将钢琴的信号转换特定的数字信号与真实的谱子进行对比纠错,从而识别错音、错节奏的现象并实时校正。这一规范技术过程被称为自动音乐转录AMT (Automatic Music Transcription)。本文采用谐波常数Q变换、CFP等不同的音乐数字特征表示方法,将原始的音乐信号转换为频谱图,作为网络结构的特征输入,改进了语义分割模型DeepLabv3+,融合了U-Net的U型结构对多乐器音乐进行转录,该算法在钢琴音乐MPAS数据集上达到了良好的识别效果。

At present, quality education is more and more valued, and music education as a representative of quality education is also more and more concerned. But music education is greatly limited by artificial educational resources. The help of artificial intelligence in music education is the process of signal type conversion from the perspective of computer. For example, for scholars to play piano, it is necessary to convert the piano signal to a specific digital signal and compare it with the real spectrum to correct errors, so as to identify the phenomenon of wrong sound and wrong rhythm and correct it in real time. This standardized technical process is called Automatic Music Transcription (AMT). The algorithm comprehensively makes use of digital feature representation methods such as harmonic constant Q transformation and CFP. It converts the original music signal into a spectrum chart as a feature input of the network structure. It improves semantic segmentation model DeepLabv3+ and incorporates U-Net’s U-shaped structure to transcribe multi-instrument music. The algorithm achieves good performance on piano music MPAS datasets.