Music Recommender ANNOY

Music Recommender using ANNOY and Google's AudioSet

Highlights

1. Predicts song similarity based on Audio Embedding Information
2. Approximate Nearest Neighbors used for feature similarity
3. Embedded data generated for every second of audio was pre-transformed with "IBM MAX Audio Embedding Generator" (https://github.com/IBM/MAX-Audio-Embedding-Generator)

How to run

1. Download Google Dataset (See references section below) and place in the ./data/audioset_v1_embeddings directory
2. Run main.ipynb

Dependencies

• Tensorflow 2.0 (installed via pip in anaconda)
• ANNOY (https://github.com/spotify/annoy)

Demo Images

Input Data gets 5 video Recommendations from YouTube

References

1. Dataset: https://research.google.com/audioset/download.html