Fall Week 10 - Project Beginnings

This week marked the beginning of the Real-Time Chord Recognition project. With my partner, I defined the project goals, technical approach, and future timeline. Simply summarized, the goal is to create a program to classify, display, and record musical chords played by a musician in real-time. Our motivation is as follows: a musician spends a lot of valuable time by writing music during the creative process, so our program will remove the burden of remembering a particular chord or chord progression, for example. By automatically, transcribing the music, the musician can focus their efforts on creating.

After doing some research, we found methods used for chord recognition that primarily make use of chroma features as the data structure used to represent chord information. Chroma features give a certain audio snippet its own harmonic signature, which is the first step towards classification. The next step is classification, for which there are several methods, but two important ones are linear classifiers and hidden Markov models (HMMs). We plan to utilize both chroma features and different classifiers based on which are easiest to implement and give the best accuracy.

I began work on the project by looking for a dataset of audio files that will be used to generate chroma features and eventually build a classifier algorithm. Some possible datasets include one constructed from Beatles and other Pop songs, or a more rudimentary collection of guitar chords. In the following week, I plan to narrow down the options and begin chroma feature processing on our selected data set.