Hi Rohin, that’s an important insight — that on a broader level, different genres of music require different modeling since 21st century pop and 1950s jazz are structured differently. At the same time, there are similar qualities among different genres as well. For instance, melody tends to depend on the harmony in general despite different genres (classical, jazz, edm, pop, etc.), as this is usually a general condition of music as a whole. It’s just that pop and EDM share this condition to a more intense degree due to the repetitive nature of the harmony. This then leads to the ultimate point about domain knowledge in data science; that it’s important for those in music generation (or any field for that matter) to have very subtle and deep knowledge about the underlying data if they are willing to based their models on it.
