Hello all,
I am about to finish the Deep learning course (I already did Machine learning and AI for everyone). I’ve done this courses just to grasp the surface of what it is possible to do and I’ve come with the conclusion that a project which I started 20 years ago could now be very useful for the music industry and ML.
The project ended around 2015 with a patented controller, at that time we loose it all (more than 100,000$ in patents, time and health) so it has been in “stand by” for some years. Two years ago I started again using my spare time, but I really need help since my knowledge is basic and it takes me days for simple tasks.
The system allows to compress a single note with alterations (up to double flat and sharps), octave, and key signature into a number (or a one-hot vector with 33 elements), preserving all properties better than, for example, MIDI.
It happens that the disposition of the notes in a matrix with fixed dimensions matches exactly the position that they would have in a staff and also the position of the notes are perfectly ordered in the space taking in account music theory.
It is the most simple and efficient way to represent written music and I my gut feeling is that it could fit in ML for most of the architectures (as a better embedding, filter in CNN, data preprocessing and so on).
As a visualization tool, it could improve “music education” for anybody.
Before showing the project I would like to know:
If anyone is interested (mathematician, musician, ML expert, developer…) in the idea.
If anyone knows about how to rise some funds (we have been already granted with 500$ that are available).
If there is anyway to talk to any of the teachers, instructors or even “Coursera” or “Deeplearning” staff itself.
Since this project is not only about ML but Music education, I think that even this platform could be interested.
Thanks in advance, and I hope someone read this topic!
I have read your topic. I have some experience with patents. I have some background in music.
I have a few thoughts, but not any significant help.
Music encoding is pretty well-plowed ground already.
Your patent will act as a deterrent to those who might want to engage in this topic with you. There is too much risk of legal entanglements later on regarding potential infringement.
Coursera is an educational course host. They do not do any technical development of their own. Their partners (like DLAI) provide all of the content.
Those who would be most likely have interest in the topic of your patent are very likely to already have suitable methods of music encoding. So finding investors will be problematic.
Perhaps I am entirely wrong. In any event, I wish you the best of fortune in this regard.
Hello TMosh,
Thank you very much for your time and reply.
Patent won’t be a problem since we are the ones who patented it and it is no longer being paid.
You are right, music encoding is already well-plowed ground but it is mostly based on MIDI and piano roll which is really good but also old technology and does not map unequivocally any note, nor does the piano. This method compress any written note from any octave with any alterations (up to double flat and sharp) with any key signature to a single number.
If you are willing to see it, let me know, I be more than glad to have your thoughts.
By the way, my cousin is the musician behind the idea, not me, he is a jazz teacher with studies in “Berklee College of Music” and also graduated in Spain (it is 14 years to complete it, he did it in 4 years after his studies in U.S.A) I am confident that there is nothing like this and almost sure that it can be applied to ML. Any way we will share the code, but I would like some second thoughts and how to approach at the very first steps.