Title Page
Contents
Abstract 8
1. Introduction 9
2. Related Work 12
3. Dataset 15
4. Method 18
4.1. Preprocessing 18
4.2. The Architecture of Pitch Classification Model 23
4.3. Experiment 26
5. Results 28
6. Conclusion 34
References 36
Table 1. The details of GuitarSet dataset with pitch annotations used in this paper. 17
Table 2. The accuracy results of precision, recall, and F1 score on test dataset after a few times of training the model. 30
Table 3. Comparison of our proposed Deep Learning Model with Previous Methods for Pitch Classification. 32
Fig. 1. A signal sample with 6 channels and annotations for each chord section. 16
Fig. 2. Chords information of all .wav files. 17
Fig. 3. The visual form of the C chord data at a frequency of 44100Hz. 19
Fig. 4. The down sampled data from 44100Hz to 22050Hz with valid features after processing using Fast Fourier Transform. 20
Fig. 5. The figure above shows the data with valid features after max pooling sampling and crop the useful range. 22
Fig. 6. The distribution of reassigned classes and weights is visualized through a bar graph, illustrating the total of 12 chord classes and the... 23
Fig. 7. Pitch Classification Model Architecture 25
Fig. 8. Training and validation accuracy, training and validation loss 29
Fig. 9. Train and Test F1 score after 200 epochs 31