Research on Cello Sound Segmentation and Note Recognition Algorithm Based on Deep Learning Networks

The primary problem of sound segmentation and note recognition in cello audio is solved by introducing a deep learning based method proposed in this paper. The characteristics of cello audio were analyzed in the time domain and frequency domain which made it possible to form a new scheme of feature extraction and data preprocessing. It was the first time that the CNN, RNN and Transformer models were used together to carry out the task of sound segmentation and note recognition. With CNNs, we managed to extract local spectral features, with RNNs, we designed Long Short-Term Memory (LSTM) units to model temporal dependencies, and with Transformers, we learned global contextual relationships, which, as a result, led to successful sound boundary segmentation and effective note category recognition. The experimental section is thorough by showing the performance of the algorithm on the dataset used in our study. We conduct performance testing and metric assessment, by which we judge the performance of the algorithm which we give feedback on. The metrics include accuracy, recall, and F1 score. Through the experiments, we demonstrate the correctness and real-time computation capability of the algorithm in both the tasks of sound segmentation and note recognition, indicating that a new technical avenue of music information processing and automatic music score generation has opened up.