1000 WAYS OF REJECTION

2020

Everyone's life has been disrupted in the year 2020. We were forced to go through more failures, rejections and negative results than usual. While a "no" represents these unpleasant situations we've been facing, it can also be a voice to protest and make a change. By creating this interactive audio-visual experience with machine learning algorithms, I'm hoping to let out the negative energy with a little humor.

After finding an audio dataset of different people saying "no", I selected 1000 and sorted them into clusters with KNN in Python. I then exported the cluster information into a JSON file and visualized them in p5.js. By moving the mouse around in the canvas, you'll hear the "no"s being spoken continuously while passing each data point.



Dataset:

Speech Commands

Resources:

Sorting Audio Clips

Using UMAP for Clustering

Plotting UMAP Results

Tools:

librosa

sklearn

UMAP

HDBSCAN