What if we had the ability to predict new lyrics of your favourite artists in their style?
Turns out, we can - with the help of Machine Learning.
Since I wanted to allow choosing any artist to predict their lyrics, I had to write a program to web scrape the lyrics of ALL songs through lyric websites. Using BeautifulSoup and python, I scraped the lyrics from websites and stored them in a pickle file. This file is the "corpus", i.e, the dataset on which I will train and test the model.
Markov chain is a concept that uses graph theory. At each node is a state - which in this case is a single word. There are connections to different states that control how the states will interact with each other.
In this particular project, I have used a bigram chain, just to ease the complexity. However, using a Markov Chain causes a certain amount of repetition (especially with a bigram model) but its still fun, and there isnt a need to do too deep into ML or NLP. Lyrics prediction can also be done using LSTM or other Recurrent Neural Networks.
My favourite artist is Hozier, and he also hasn't released any new music. So naturally, I tested this with his style, and the lyrics came out pretty good:
