Farewell - Artificial Pianist

We were teammates Batuhan Ayvaz , Batuhan Orhon and me, we work on the project Artificial Pianist within the course of BBM406-Fundamentals of Machine Learning throughout this semester. We want to thank our professor Erkut Erdem and our teaching assistant Necva Bölücü for their guidance.

The whole process of developing the project was very enhancing and helped us to be involved in the artificial intelligence field in various aspects, like writing blog posts, reading research papers, and following the latest improvements.

We should say that it was a little complicated at first for us who are beginners in machine learning, but at the end of the project, I mean the whole process of researching, learning, trying, we can see how much we improved ourselves and gained knowledge.

We have made a user study for the 3 music pieces generated by different model architectures and collect points between 1 to 10 from listeners and calculated mean opinion score for each piece as subjective evaluation.

Blossoms is one of the first music pieces that has generated by using LSTM layers and self attention. Blossoms got 5.7 points from listeners in average and this is the lowest score. We received comments about Blossoms like it is atonic and most of the notes are just random, it has less harmony and not euphonic, some parts of the music are repeating often.

After the generation of Blossoms (and the other similar music pieces), we have decided that we need further improvements on the architecture of model and we have changed the model using Transformers. Results of this model were explicitly better than the first results. Raven also has been generated by using this model, and got 7.3 points and received nice comments from listeners like it is more harmonic, soft and less atonic.

Lastly, we have tried using dropout layers in the model. Yellow has been generated by using this model. Yellow got 7.5 points from listeners and nice comments too. There is not a big difference between the music pieces generated by the second and third model but we can say that the third model is slightly more euphonic.

We also wonder about your comments, which one is more euphonic to you ? Feel free to comment whenever you read it.

And we have come to the end of the blog post series of our project Artificial Pianist and we want to say goodbye.

Goodbye !