Songbeast

TLDR: Songbeast is the Rotten Tomatoes of music.

Songbeast is an AI driven song analysis hit predictor. Music artists and label A&Rs can input an unreleased song (pre-release) and determine if it's viability to be a hit by comparing it to 1.25M popular songs released within the last 5-6 years. Songbeast is a subscription based platform. Users can rang from novice creators to label A&Rs. Songbeast's AI has already been built and stress tested. We have self-funded this project up to this point. All invested funds will be used to continue the AI learning, building out the website itself, and marketing it.  

We are a team comprised of two. As founders we both have had individual success as creators in the music industry. We have a combined 20+ years of experience and we both have been signed to a major labels and publishers. As musicians ourselves,  we both intimately understand the key elements that are important when it comes to music and machine learning. 

Songbeast uses superior data for training.  Other attempts at hit song prediction machine learning models have relied primarily on data from Spotify, but this may be a mistake.  The #1 music streaming platform in the world is YouTube, by a large margin (approximately 350 million monthly music listeners on Spotify vs 1-2 billion for YouTube. More songs are streamed on YouTube than Spotify, Apple Music, and every other music streaming platform combined). 

 

In machine learning, nothing is more important than the data, and YouTube has Spotify beat.  YouTube is the #2 most popular site in the entire internet, and music streaming is the primary activity of YouTube users. Critically, the song stream data from Spotify is questionable.  Spotify is notorious for fake streaming numbers, which can be produced by bots.  Artists and record labels ranging from small-time to big-name have been implicated in purchasing fake streams to boost their numbers to increase their royalty payments and rankings. 

 

 

Songbeast has a superior methodology if the user is looking to predict if a song is going to be a hit or not, based upon current musical taste, which is its intended use.  Some hit song prediction models may be trained on songs going back decades.  SB was exclusively trained on songs ranging from 6 months to 5 years old, so it’s only trained based on current musical tastes.  The model was trained to be conservative in labeling a song a hit, by making the “Hit” threshold a lofty 50 million YouTube views (50 million and above is classified as a hit, under 50 million is classified as a flop), so users can be confident the model has high standards for what qualifies as a hit.  The testing, training, and validation of the model encompassed a total of approximately 1.25M songs.

 

Another key difference in methodology is the fact that Songbeast bases its entire prediction solely on the music, while other hit predictors include the artist’s past history in their prediction.  This can be considered a cheat that greatly undermines the true accuracy of other prediction models, because if you’re dealing with a huge artist with many hits, the model will almost always predict their songs will be a hit.  If you’re dealing with an artist who has never had a hit, the model will know this artist has never had a hit, and will likely predict all songs will be flops because of the artist’s past history.  Both of these extremes miss the most important thing, the music itself.  Our model makes predictions solely based on the music, not the past history of the artist.

 

Songbeast is trained on the best available music data, trained purely on music without artist info, has high standards for what qualifies as a hit, and maintains a focus on current musical tastes.  

Imagine seeing the Songbeast logo (with score) next to every released song on every streaming platform. Songbeast scores will be as integral to deciding whether to listen to new music as Rotten Tomatoes scores are to potential movie-goers. Imagine every bedroom pop producer and artist in the world able to compare their work to the most popular current music in a quantifiable and accurate way.