Week 1 — Tune It Up
Hello world,
We are Fidan Samet, Oğuz Bakır and Adnan Fidan. In the scope of the Fundamentals of Machine Learning course project, we are working on prediction and style transfer of song release years. We will be writing blogs about our progress throughout the project and this is the first one of our blog series. The problem we consider, our motivation, and the datasets we may use will be covered in this post. So let’s get started!
Problem & Motivation
“Music is the food of soul.” said Arthur Schopenhauer. He was quite right. Music nurtures the soul and expresses feelings that cannot be expressed in words. It is an art that appeals to everyone. People find themselves in music, so music becomes a necessity for the soul. The below figure is the chart of album sales per year. It shows that even if the physical state of the albums changed over time, the album sales remained above 500 million.
As technology improved, people preferred to stream music instead of buying albums. The below figure is the chart of Spotify's growth timeline. It shows that people feed their souls with music as was the case decades ago.
When talking about music that has such an impact on human life, it is impossible not to mention Machine Learning that adds many innovations to the art domain. Various Machine Learning applications on music, such as music identification and recommendation, have been and are still being developed. In these studies, many models on song classification and style transfer were developed. What about predicting and transferring styles of song release years? That sounds interesting.
Over the decades, styles of songs changed, the usages of electronic instruments and auto-tune have increased. Just to observe these changes, let us take you back in time a little bit.
The 90s are not so far:
Sure 80s Axl Rose would feel the change:
What about the 70s:
The Beatles from the 60s:
The 50s are far away:
Along with these changes in music, we decided to work on a task that would be interesting and challenging. We aim to perform classification and style transfer of song release years. Doesn’t seem easy, does it? We plan to use deep learning methods to realize these tasks. Our initial aim is to use lyrics, timbres, and audio signals of the songs as features. Of course, we can consider using different features but we will explore that together.
Datasets
There are several existing song datasets in the literature. We consider using the following datasets:
We will prepare and use subsets of these datasets according to our tasks. Topics related to the datasets will be covered in-depth in future blogs.
That is all for this week. Thank you for reading and we hope to see you next week!
