`` Maroofy '' that AI finds `` songs with atmosphere similar to favorite songs '' from among 120 million songs

Even if you like a certain song and think 'I want to hear more songs with a similar atmosphere!', it is extremely difficult to find songs with a similar atmosphere from music streaming sites and video sites using titles and album art as clues. is. Therefore, programmer

Subhash Ramesh has released an AI service `` Maroofy '' that finds `` songs with a similar atmosphere to a specific song '' from 120 million songs in the iTunes Store.


When you access Maroofy's top page, it looks like this. A search form is installed in the center, and the recently searched songs are displayed below it.

As a test, search for '

only my railgun ', which is the OP theme of the TV anime ' Toaru Kagaku no Railgun '. When you enter a song name in the search form, candidates are displayed in a row, so find the appropriate one and click.

Then, similar songs are displayed in a list, and when you click the play button, the sample part registered in the iTunes Store will be played. In addition, there is a tendency for different versions, cover songs, and remixes of the original songs to be displayed at the top of the list, but songs by unrelated artists are also lined up.

If there is a song title that you are interested in, you can click the play button and check if it is actually similar. Looking at the list, there was a song '

Nanchu Koi wo Yatteruu YOU KNOW? ' released by Berryz Kobo in 2005. Listening to the sample part, I feel that it is indeed similar.

When I searched for the main theme of the movie ' Star Wars ', the main theme of the movie ' Back to the Future ' was displayed at the top. When you compare them, the magnificent performances by the orchestra that heighten your anticipation for the story will make you feel similar emotions.

Maroofy said that 120 million songs in the iTunes Store were input as raw data and indexed using a custom AI audio model. Using this database, it seems that you can find similar music by semantic search.

In addition, Hacker News, a social news site, has a thread set up by Mr. Ramesh himself, soliciting feedback from users. One user said, ``Although songs with similar sounds are selected, there is no consistency in tempo, genre, vocal tone, etc., and it seems that the similarity between songs is selected only in the sample part.'' While saying negative points, we have received feedback that it would be useful if we could filter the search results in some way. In addition, there are positive comments such as 'It's better than Spotify and YouTube Music, which only recommend popular songs', and 'It would be even better if you could filter amateur songs'.

Show HN: I trained an AI model on 120M+ songs from iTunes | Hacker News

in Review,   Web Service, Posted by log1h_ik