Greetings everyone! I’m Peter Li and welcome to my second blog! Here are some updates on what happened this week and what I hope to accomplish next.
I started my research process by entering the world of GiantMIDI, a project developed by the technology company ByteDance. Founded in 2012, it is now one of the biggest companies in the areas of software and artificial intelligence. (It is also the developer and owner of TikTok!) As of right now it is comprised of two main parts: one being a transcription algorithm, which takes piano recordings as input and returns transcribed MIDI files, and the other being a large database containing music score, recording, and more resources ready to be transcribed by the algorithm. (Note: MIDI stands for Musical Instrument Digital Interface. It is a file type that electronically stores musical arrangements, and is the primary way electronic music is created and stored.)
The main thing I did this week was reading published research on GiantMIDI, while also doing some coding with the algorithm I mentioned in last week’s blog. According to a research paper published by GiantMIDI’s development team, they have conducted large-scale statistical analyses of the database over several different variables. For example, they have produced a “notes histogram” of various composers, shown below.
This is significant in that it demonstrates the different ranges composers of different eras tend to compose within, which matters when developing software that can create music that mimics a certain time period/style (such as Baroque, Classical, Romantic and Impressionism). Notice in this example where as the composer gets more modern (from Bach to Beethoven to Liszt), the frequency of the lowest and highest registers increases. Trends like these are what music composition software rely on to get a sense of what different genres/eras of music share and differ, which I will explore more in the coming weeks by experimenting with the transcription algorithm.
I will keep posting new findings as I continue to run and analyze the program and published research associated with this project. Meanwhile, thank you for reading my blog! See you next week, and stay tuned!