These days, there is no platform or service that doesn't try to recommend "something" for us to watch, listen to, or buy. The problem is that in most cases, these recommendations are truly horrible. So why don't we take control a bit more? The Maroofy engine points in that direction with music suggestions based on favorite songs we previously search. This way, the results should be more accurate…
No matter if it's YouTube, Spotify or another similar service, the "suggestions" from all modern platforms are constant. Why? Because the wheel needs to keep spinning, of course. Big Data doesn't feed itself: we must spend time online, trying new or classic things, consuming. But the algorithms are wrong… much more than we imagine.
Maroofy: Discover New Songs by Sharing Your Favorites
Now, there is a simple alternative: voluntarily teach the algorithms what we want. That's how Maroofy works. This site, developed by Subhash Ramesh, offers us a "recommendation engine" for similar songs. Its operation is based on an index with more than 120 million songs on iTunes, and a custom audio model.
The model analyzes raw audio as input, and produces an output based on vector embeddings. The last step is to store all those vectors in a database, and enable semantic search to find similar music.
So, how well does it work? Honestly, it depends on each person. Personally, I can say that currently the "discovery" factor of Maroofy is more valuable than everything else. For example, the engine indicates that Fortunate Son by Creedence, a song "protest, anti-war, and anti-elitist", shares similarities with Grease by Bummertrash, an indie band of which I could barely find their Bandcamp profile and that is more in line with garage punk than anything else... but in the end I stayed listening to Bummertrash, so, mission accomplished? Try it!
Official site: Click here
