ISP – intelligent song picker

Have you ever felt overwhelmed when choosing a song from Spotify?

This is the moment when ISP will help you.

All you need to do is take a selfie. ISP will do the rest of the magic* and provide you with a song that fits your musical taste, feeling and environment.

Don't worry, the picture stays on your device.

_{^{*Machine learning for face recognition, emotion classification, fancy song search algorithms and other lovely stuff}}

Demo

Try now

💥 Try it out right now at intelligent-song-picker.com!

Background

I developed this project during the course Interaction/Programming by Prof. Dr. Thomas Rist at the University of Applied Sciences Augsburg. In addition to the web application I wrote a paper about this project.

Setup

Create a local running copy of this project

Client

1. Install dependencies

   cd client
   yarn install
   

2. Start local development server

   yarn serve
   

3. Or create a build for deployment in production

   yarn build
   

Server

1. Install dependencies

   cd server
   yarn install
   

2. Run local nodemon server with hotreloading

   yarn serve
   

3. Or run production server on node only

   yarn production
   

Find out more

This is a list of the core dependencies used in this project for your further reference.

Client

• vue.js
• tensorflow.js
• tfjs-emotion-classification (not a "real" dependecy but implemented)
  • Blazeface detector
  • Emotion classification (CNN Model)
• webpack

Server

• Node.js
• Express.js

ToDo

• Improve stability
• Add Autoplay function to go on playing songs after the first one is finished
• Handle Spotify non-premium users
• Add contenthash to js files for improved caching
• Create robust mobile version (currently a very bad workaround is in place here)
• Reduce own queries to spotify and instead leverage the capabilities of the Spotify Web Playback SDK