Embeat

Embeat is an open source music recommendation system from GD Studio, built on Spotify acoustic feature data. Given a seed track, it returns songs that sound alike and songs that listeners tend to pair with it, combining a contrastive learning model for acoustic similarity with a collaborative filtering model learned from millions of playlists. It powers the similar songs feature of GD Studio's online music player, GD音乐台, and is published on GitHub with model weights and datasets on Hugging Face. It suits developers and hobbyists who want to run their own music recommendations rather than rely on a streaming service. EmbeatMLP encodes audio features such as key, tempo, energy and mood into 64 dimensional vectors, while Track2Vec captures playlist co occurrence patterns. Micro genre tags covering more than two million artists keep results stable for niche songs, and five recall channels (acoustic similarity, popular tracks in the same genre, same artist, similar artists and playlist collaborative filtering) are merged and scored for the final list. Seed tracks can be looked up by Spotify track ID, ISRC, title and artist, or artist alone, and the project documents deployment on a VPS with 2GB or more of RAM using a Qdrant vector database. Embeat is free to download. The code is released under the Creative Commons Attribution NonCommercial 4.0 license, so commercial use is not permitted, and the 45 million track dataset, model weights and prebuilt databases are also published at no cost. The listed website is GD Studio's music player rather than a dedicated Embeat page. That player aggregates public interfaces from multiple music platforms and carries a disclaimer limiting use to personal study, so buyers evaluating Embeat should work from the GitHub repository instead.