NextSong- Smart Music Recommender Using ML

Team: Michal Moryosef, Abel Asfaw, Dor Ulman, Tenzin Sherpa

Welcome to NextSong! smart music recommender, using machine learning.

Demo

https://www.youtube.com/watch?v=_EVDI73n0sU

Jupyter Notebook + HTML View

Included in this repo inside Jupyter Noebook folder. If for some reason you can not access to it please find it here: https://drive.google.com/file/d/14vR92W6vfm4MUqNBWYZDgtj_hZXuvvLP/view?usp=sharing https://drive.google.com/file/d/1Y5u6gnrgd5_Ts83fuJUs21UBriU5GwLN/view?usp=sharing

*In order to view visuals in HTML format you need to download Music Files folder (in Google Drive link) to the same path of your HTML.

NextSong Project Deck Slides

https://docs.google.com/presentation/d/1vv_OHbvcxyG1vmFbQ4ac_xXvftjzGMk01B1HucKkwiM/edit?usp=sharing

Brief Background:

The goal of this project is to study how music affects heart rate variability and see how audio features affect our heart rate speed. This has both entertainment and health benefit.
In the development of the product, NextSong used:

1. "CooSpo Heart Rate Monitor" to monitor heart rate.¹
2. Pulsoid - heart rate widget for live streams used to broadcast user's live heart rate to NextSong server and feed it later to the model.²
3. Spotify API for all music analysis ³
4. "musicnn"- an open source, deep learning-based music tagger, used as NextSong tansfer model ⁴

¹ https://www.amazon.com/gp/product/B07R8741CN/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1
² https://pulsoid.net/
³ https://developer.spotify.com/console/
⁴ https://github.com/jordipons/musicnn, https://towardsdatascience.com/musicnn-5d1a5883989b

Technologies Used:

1. Flask
2. Bootstrap
3. Methods and Algorithms: CNN, Linear Regression, Transfer Learning
4. API's extensive usage from Spotify and Pulsoid

Pre-Requisites To Run NextSong Locally:

1. Clone project

Running the pretrained model:

You have to create a virtual enviorement for this, we recommend using anaconda. Once you have anaconda installed on your machine: (https://www.anaconda.com/products/individual), open anaconda and then run the follwing:

Conda create —name myenv
Conda activate myenv
After that run the requirements txt

• Make sure python is 3.7 not 3.8

2. Install the necessary Python packages

$ pip install -r requirements.txt>

3. Own a Spotify account to get the music recommendation playlist created directly to your account.

https://www.spotify.com

4. Export the environment variables

$ export SPOTIFY_AUTHORIZATION_TOKEN=value_grabbed_from_spotify

$ export SPOTIFY_USER_ID=value_grabbed_from_spotify

5. Have a Pulsoid app account in order to live stream your heart rate to the website

https://pulsoid.net/

6. In cloned project folder run the entry-point script in cmd/termial

python app.py

7. Download "musicnn" transfer model from GoogleDrive link below and put in your cloned folder

(We couldn't attach it directly in this repo due to size limitations) https://drive.google.com/drive/folders/1H9v07PqVAwPicbWGRxhDk3EQb3TY9VnF?usp=sharing