Group5-CMLS-HW1

Project Workflow

Preprocessing

• retrieve annotations
• check original dataset
• split the dataset into training/test (80% vs. 20%)

Observe the Dataset

• have a look at the data
• check the quality

The Features 

• 34 columns of feature
• 400 audio tracks
• 80 % for training
• 20 % for testing
• rescaling, selection, saving

Two Classifiers: KNN & SVC

• The process to use the classifiers: pick the right parameters --> build the classifiers -->fit with training data set
• We use K-fold (k = 10) Cross Validation and Grid Search method to optimise the parameters for those two classifiers

Results & Conclusions

• Here are two confusion matrix to represent the final testing result, separately for  KNN and SVC

• Each row in the matrix is the real genre, while each column is the predicted genre category classified by the model we build

Plot 4. KNN & SVC results of Confusion Matrix

Table 3: Final Results

Reference

• [1]Data Sets. GTZAN Genre Collection.url:http : / / marsyas.info/downloads/datasets.html.

• [2]LibROSA Documentation.url:https://librosa.github.io/librosa/.

• [3]NumPy.url:https://numpy.org/.

• [4]Pandas documentation.url:https://pandas.pydata.org/pandas-docs/stable/index.html.

• [5]Scikit-learn.url:https://scikit-learn.org/stable/.

• [6] Bob L. Sturm. “The GTZAN dataset: Itscontents, its faults, their effects on eval-uation, and its future use”. In:CoRRabs/1306.1461 (2013). arXiv:1306.1461.url:http://arxiv.org/abs/1306.1461.

• [7] G. Tzanetakis and P. Cook. “Musicalgenre classification of audio signals”. In:IEEE Transactions on Speech and AudioProcessing10.5 (2002), pp. 293–302

• [8] Final Detailed Report, Music Genre Classification, https://yilin10.github.io/MusicalGenreClassification/final.pdf

• [9] Final Google Colab IPython Notebook, Music Genre Classification, https://github.com/yilin10/MusicalGenreClassification/blob/master/Assignment1.ipynb