Week 4 – Getting Theoretical?

Apr 02, 2021

What’s good everyone! My name is Peter Li and welcome to my fourth blog! I can’t believe it has been almost a month since the start!!

I have been continuing the first part of my project by expanding my research into the algorithms and papers by ByteDance’s GiantMIDI development team. As I mentioned last week, I realized the importance of music theory in analyzing the music database and creating music composition software, which is what I will (try to) explain right now. 🙂

Two figures from the statistical research and analysis done by the GiantMIDI team are shown below regarding the average occurrences of various trichords and tetrachords. (The difference between them is that tetrachords contain four notes instead of three. They sound more “full” than trichords and can have more combinations, which can create unique harmonies that trichords cannot provide.)

As shown, the frequency (number of occurrences, not pitch 😂) of six common trichords and tetrachords each were examined in the works of six famous composers, each representing an era/style of music. My main takeaway from these graphs is that though the sample size is relatively small (only 12 chords examined), they somewhat reflect what harmonies composers tend to use the most in each era, which provides great help to AI composition software in creating music that mimics a certain style. For example, Mozart and Beethoven, being Classical composers, share similar graphs, while Chopin and Liszt, being Romantic composers, have similar graphs that are different from the Classical ones. One interesting note in my opinion is how the graphs of Debussy show that the occurrences of most chords examined in his works are less than the other 5 composers, which are all not in his era. I think this does not mean that he has less works or that his works contain less notes – to me it is that he, representing Impressionism, tends to use very different harmonies and chords that are not shown in these graphs. This shows that although these results might be useful in helping machine learning to create Baroque and Classical-like music, we probably need more than this for other styles including Impressionism and Modernism.

I have an important update concerning another big part of my research process coming in the following weeks, so stay tuned! And as always, thank you for reading & I’ll talk to you in the next one!

3 Replies to “Week 4 – Getting Theoretical?”

  1. Leo L. says:

    Hi Peter your progress looks great! I like the anlysis on frequency. Can’t wait to see more results!

  2. Dora X. says:

    Your research looks really interesting! I’ve never seen analysis on chord frequencies before so it’s amazing how differences in harmonies can represent different eras or styles of music. I look forward to seeing more!

  3. Jiaming Z. says:

    Bruh how did you add emojis to your post? This is very interesting and cool, although I don’t know much music, but still, very cool :)))

Leave a Reply