A Python-based audio classification project for CPSC 473 - Introduction to Data Mining

• This program requires:
  • Librosa, a Python package "for music and audio analysis" (developed using Librosa 0.8.1)
  • Pandas, a Python package for data analysis (developed using Pandas 1.3.4)
  • Scikit-Learn, a Python package for data mining and machine learning (developed using Scikit-Learn 1.0.1)
• All of the packages list above, as well as all of their dependencies, can be installed using the package manager pip by simply executing the following shell command: pip install -r required_packages.txt
  • Note that the shell instance in question must be navigated to the root directory of this project (i.e. where required_packages.txt is found)

• To run the program, simply run main.py, either using the command python on the command line (while in the same directory as the source files), or the IDE of your choice
  • Instructions regarding choice of pre-processing techniques, etc. are all provided in-program
  • If you wish to skip preprocessing the .wav files (A very lengthy process as per .\preprocess_times.txt), preprocessing has already been performed using all 3 methodologies, and the outputs have been saved
    • Simply copy and paste the files found within .\wavDataPP0\ (or \wavDataPP1\, or \wavDataPP2\) into .\wavData\, where the X in PPX indicates which of the 3 preprocessing methodologies the contents were generated using (see the results section in the project paper - .\CPSC473_project_paper.pdf - for more clarification)
  • Alternatively, if you would like to preprocess the .wav files yourself, they can be downloaded via the URL found in .\wav_source.txt
    • The .wav source used is the UrbanSound8k dataset; all compilation credit goes to Justin Salamon, Christopher Jacoby and Juan Pablo Bello. The .wav files included originate from www.freesound.org.
    • After filling out a brief download form, you should be provided with a tarball. Within, there should be folders containing .wav files named fold1 through fold10 (.\UrbanSound8K\audio\fold1 ... fold10\). Place these folders directly into the project subdirectory .\wavs\.
      • Thus, the project subdirectory .\wavs\ should, in turn, have subdirectories .\wavs\fold1 ... fold10\

• Once preprocessed, data (CSVs) for each .wav for each of the ten (10) folds of .wavs can be found in .\wavData\ in files .\wavData\fold1Data.txt ... fold10Data.txt
• Expected classifications (ordered) for each .wav within the designated test fold can be found in .\ExpectedResults.txt (following a run of the program)
• KNN and Logistic Regression model output (classifications) for each .wav within the designated test fold can be found in .\KNNOutput.txt and .\LogRegOutput.txt, respectively (following a run of the program)