irfaan / song-embeddings

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Leaning song embeddings with Doc2Vec

The code in this repository was used to learn embeddings for songs using their lyrics. Training data consists of 193 songs by the following artists:

  • Blink-182
  • Angels and Airwaves
  • Box Car Racer

Using the embeddings for playlist generation

The following steps were be taken to create this spotify playlist

  • A Doc2Vec PV-DM model was fit to the corpus of songs
  • Song embeddings (high-dimensional vectors) were predicted for each song
  • A similarity score was calculated between each pair of songs
  • A spotify playlist was created by starting with one song, then finding to its next closet song that is not yet in the playlist

About


Languages

Language:Jupyter Notebook 100.0%