Deezer's AI Detects Mood, Makes Your Playlist Smarter

Is an artificial intelligence smart enough to match up music and moods? According to Deezer it can be, as the music streaming service developed a new AI capable of detecting a song’s mood and responding accordingly. By using this new AI, Deezer can in theory provide more accurate playlists and song sorting specific to a subscriber’s mood.

This could result in Deezer suggesting tracks that make you feel happier without necessarily resorting to cheesy pop, or tracks that can chill you out without you drifting off. It could also be used to look into the relationship between music, lyrics, and mood, as well as being able to sort through lots of unlabeled data. The AI then assigns tags to new songs based on their audio information and lyrics.

First spotted by Venturebeat, the research behind the artificial intelligence is explained in a new paper called “Music Mood Detection Based on Audio Lyrics With Deep Neural Nets”. The paper describes how the AI was trained by combining Deezer’s music catalogue with the Million Song Dataset (MSD) – a database of tracks that uses LastFM tags, some of which have mood information attached. These tags were assigned values based on two scales: one for the valence (from negative to positive mood) and one for the arousal (from calm to energetic mood) of an emotion.

Since the MSD doesn’t include audio signals or lyrics, the data from it was combined with Deezer’s song metadata (including song title, artist name, and album title) and lyrics. As the paper explains, “for a certain audio segment, we extract words from the lyrics that are at the corresponding location relatively to the length of the lyrics”, although the data about the lyrics cannot be released due to rights restrictions. Approximately 18,000 tracks were used to train the artificial intelligence, with the researchers concluding that the AI was more capable in detecting whether a song was calm or energetic in comparison to “classical approaches on arousal detection”, and performed about the same when it came to valence.

The system isn’t perfect yet, though.

The researchers point out that if they had a database of labels indicating ambiguity of a track’s mood, their results could be more accurate.

Not everyone can agree whether a song is positive or negative, so having access to the variety of opinion would be particularly helpful.

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *

       
Apps & Games Clothing Electronics & Photo Large Appliances
Baby Womens Apparel Garden Lighting
Beauty Mens Apparel Outdoors Luggage
Books Girls Apparel Health & Personal Care Pet Supplies
Car Boys Apparel Home Shoes & Bags
Motorbike Computers & Accessories Kitchen Equipment Sports & Outdoors
Fashion DIY & Tools Jewellery Toys & Games