AI Vocal Remover

Inspiration

The theme reminded us of a time recently where we heard a song with an instrumental we really liked, but could not find just the instrumental, so we decided to create an application that utilizes machine learning to make the instrumental file for us.

What it does

Our application takes a WAV file from the user and eliminates all the vocals in the file. It outputs a WAV file of just the instrumental and a WAV file of just the vocals

How we built it

We implemented our project using python and the pytorch machine learning framework. We then trained the neural network we used from github user tsurumeso. While training we implemented a GUI to make the program easier to use.

Challenges we ran into

Implementing a GUI that seemlessly worked with a complicated architecture.

Accomplishments that we're proud of

Creating a project that is useful and interesting and relates to our passions.

What we learned

Both of us are fairly new to machine learning, so we learned a lot about how complicated architectures are formed and how they work. We also learned a lot about the intricacies that go into training a neural network.

What's next for Vocal Remover

To improve Vocal Remover, we would want to train the neural network for a longer period of time with a bigger dataset of songs. We would also like to fix the bugs that occur when using the postprocess option.