Sal 220c
Welcome to my 220C Project Blog
April 10, 2023
Today, I am demonstrating the type of music I want to create throughout the quarter. My first project is a demo of a "(I Can't Get No) Satisfaction" remix that I made last quarter. Unclear if I will continue to develop this project or move on to something else. My current end goal for the quarter is to have 1-3 well produced songs (originals or remixes) to show to the class. Maybe I will bring my CDJs and do a short performance that includes the songs and some other mashups.
Click here to listen [1]
April 12, 2023
Today, I am demonstrating riffusion, a new app that can interpolate between spectograms to mix music. This week, I cloned the repo and was able to run riffusion locally with full functionality including audio-to-audio style transfers, spectogram generation from natural language, and - the part that will impact my project most - the stem extractor. I used the stem extractor to split AC/DC's TNT, which I am hoping to remix as part of my spring rock remixes.
April 17, 2023
Today, I made lots of progress on my TNT remix. I was able to pull and clean the stems, and format the first half of the song. I designed the bass and worked on some synth sounds for the drop. I need to refine a lot of it, but I have reached a point where the samples are ready and structured.
April 19, 2023
Today, I decided that I will expand the scope of my presentations. I am going to work with the Riffusion founder on a computer vision project that I will reveal throughout the spring as I also work on my creative project. The main goal of this class is still to complete my creative project, but I figure that the class will also be interested in the Music/AI crossover research that I am engaging in. I am reading a number of music-ai research papers and choosing a research project next week.
April 24, 2023
Today, I presented an update on my CS231N project (which is done in conjunction with this class). I am going to try upsampling generated audio files with spectrograms. See the full project proposal here [4]. I also presented a first version of my TNT remix which goes up to the first drop. This was the first time I have shown people my music outside of my friends and it went well.
May 3, 2023
Today, I demo'd the first part of my second song and showed some upsampling projects. NuWave2, HiFi Gan, Noise2Music, and WSRGlow are all leading upsampling models who can demonstrate super resolution. I am meeting with the riffusion people at the end of this week to decide on an upsampling model architecture. Finally, I am going to meet with Chris' highschoolers to talk about a data labeling project that they might be able to help with at riffusion.
May 10, 2023 Today, I have presented an update on the creative side of my project. I have started to produce an original VIP that should be a high energy festival track. I have up to the first drop and need to work on drop variations. I will be collaborating with someone on this track and in return I am helping my collaborator with his visuals.
May 15, 2023
Today, I presented a baseline for the upsampling project. One version will be audio to audio upsampling and the other will be specrogram to spectrogram. I have created a downsampler to create my own dataset and am working towards creating a novel model and using existing models. Existing models exist in audio to audio form. My model will include spectrogram only implementations.
May 22, 2023
Today, I presented an updated version of my 'original track' which is now become a Chief Keef "I don't like" remix. I am using the vocal from Kanye's Don't Like.1 remix. Specifically I am using the first two verses from Pusha T and Kanye. My track is written in F#m and the Don't Like.1 track is in the same key. I am playing around with changing the chord progression of the first break / melody to match the chord progression of the Ye track.
FINAL UPDATE
Here are the tracks I presented in the CCRMA concert!
Original Track #1[5],
Enter Sandman Remix[6]