Predicting Chords with Machine Learning

About the project

In this project (see 'project' directory above), we use modern machine learning skills to build a model, and try to predict the chords of a given song. The input data (data fed into the model) will be the Fourier Frequency Amplitude extracted from the training data (see the .xlsx file) using Fourier analysis. The output (prediction) of the model will be chords, which will be transformed into a numeric array.

See slide introduction of the project : ccClub - Chord

---

Prerequisites

This project is mainly written in Python, recommended version between 3.6.x~3.9.x

1. Please check whether you have ffmpeg

2. It is highly recommended to use Juperter Lab (or Jupyter Notebook) to run this project. To install,

   python -m pip install jupyterlab openpyxl
   

3. The packages we need is written in requirements.txt, to install,

   python -m pip install -r requirements.txt
   

---

Get started

1. Open file ccmain.ipynb

2. The first time you run ccmain.ipynb will download music data (.wav) from the internet, it may take a few minutes.

3. Change the hyperparameters of the model, or try different structure.

4. Predict the chords of a song with the trained model.

---

Python Scripts Description

• ccdownload.py : Downloading video from the internet and transform into .wav format using ffmpeg.

• ccplay.py : Read the wave array and play the music with packages Librosa, Sounddevice.

• ccchord.py : Anything related to chords, e.g. transforming chords into numeric numpy arrays, filtering chords we need.

• ccpreprocess.py : Preprocessing the data we need, such as using Fourier Analysis to extract the wave amplitude with some target frequencies.