Researchers at Deezer developed an artificial intelligence system (AI) capable of linking songs to moods and interests. The work was described in a recently published paper on Arxiv.org, "Discovering Music Moods Based on Deep Sound Neural Networks."
To determine the mood of the song, the group considered both vocal and song lyrics. At first, they fed audio signals in a neural network, along with models that reconstruct the language contexts of the words. Then, to teach him how to determine the mood of the song, they used Million Song Dataset (MSD), a collection of metadata for more than a million songs Contemporary.
In particular, they used the Last.fm data set, which defines definitions for audio clips of more than 500,000 unique tags. Many of these ratings are related to mood, and more than 14,000 English words of these ratings have been given two levels of assessment commensurate with the negative or positive side of the word, as well as the degree of quiet or activity of the word to train the system.
The Million Song database only contains rating data for songs, not the songs themselves, so the team associates all of this information to the Deezer list using identifiers such as song titles, artist names and album titles. About 60% of the resulting data set (18644 songs) While the remainder was used to validate and test the system.
The researchers concluded in Deezer that artificial intelligence was more able to detect the quietness or activity of the song better than the traditional methods that did not use artificial intelligence, as well as when it comes to finding out whether the song is positive or negative. "These gains in performance seem to be the result of our model's ability to reveal and use the connections between sound and lyrics," the researchers wrote in the paper.
New partnership between Deezer and Amazon Alexa to facilitate listening to music
It is noted in the paper that in order to benefit from this work, "the database with the words and sounds of songs will greatly help to move forward." If there is such a database, the team believes it can determine the ambiguity of the mood of the songs. "In some cases, there can be a significant difference between listeners. People may not always agree on whether the song is positive or negative, This type of work is useful as a way to further consider how music, words, and mood are related.
Other useful links:
Deezer researchers developed an artificial intelligence system that detects the mood of songs
Reviewed by abde
on
octobre 28, 2018
Rating:
Reviewed by abde
on
octobre 28, 2018
Rating:
Aucun commentaire: