Week 1 — Introducing Our Project : The Artificial Pianist
- Project Introduction
In this project, we are going to implement a deep learning model which trains some given midi files and creates its own song. Some authorities think that in the future most of the musicians are going to use that kind of AI tools to produce their art. AI Pianist is a deep learning project which processes given audio data and learns a model based on the data. After that, AI Pianist produces its own series of notes (new song). Another feature that makes AI pianist interesting is that, when it is given a part of a song, let’s say the first two minutes of a three minutes long song, it can complete the rest of the song by himself based on the model it learned.
- Implementation Plan
At the start, we need to decide how are we going to process audio. Midi files seem desirable to use in such a project. So, we found a midi file dataset “The Maestro”.
Check ->The MAESTRO Dataset (tensorflow.org)
Pretty_midi library seems usable for the audio processing. Format of a midifile for the pretty_midi library is shown below.
In pretty_midi library there are some specific functions for the piano instrument and they are also going to be useful in the next phases of the project.
- Topics we are going to cover in next week
- Technical details of preprocessing?(Data Structures, Data Framework, etc.)
- Which deep learning model to choose for the base?
- Which additional features we are going to implement?
Stay Tuned!
Project Member
