Strawberry Beats

Inspiration

Pop culture inspired us, in our eyes, in our humanities classes that culture is built, and the core tenet that the ideas of today in music are inspired from the ideas of yesterday. We also acknowledge that people are often able to create more coherent and purposeful message by mixing up songs with intent. As a result, we focused on the ability to generate chords, lyrics, and the ability to mix up songs.

What it does

Our mashup tool allows users to input mp3 files to be able to splice, edit, and move each part to create their own mixes. Our chord generator generates a chord based off the mood. The random lyrics are generated based off the seed or topic to start off the generation.

How we built it

The mashup tool was built through wavform.js and webaudio API. The website is based off HTML. The chord generator was explicitly written through mapping the mood to a chord series based off our background music information. The lyric generator was built through a LSTM neural network in Tensorflow, which was the predecessor to LLMs in language generation. We chose a LSTM because we wanted to fine tune the result to our data. It works through tensorflow as a Recurrent Neural Net to train itself.

Challenges we ran into

The LSTM ended up breaking due to some changes that emerged from attempting to optimize the model, but it broke otherwise. Unfortunately, it broke after we were able to run it and achieve some results (Though the quality of the results might not be as accurate as we hoped). The HTML to Python file connection was something we weren't able to figure out in time unfortunately. We also chose an inefficient method of gathering data that ate up much of our time, and the model that we chose was not as coherent as a LLM.