Deezer researchers have developed an AI system that associates specific songs with mood and strength Venture beat. This work is published in "Detection of musical emotions based on lyrics using Deep Neural Network" recently published at Arxiv.org.
To determine the mood of the song, the team examined both the audio signal and the lyrics. First, we introduce a speech signal into a neural network and introduced a model to reconstruct the linguistic context of words. Then, in order to teach how to judge the mood of the song, they used the Song Million Song (MSD) dataset, which is a set of meta data of over one million songs of contemporary songs. . In particular, I used the Last.fm data set. It assigns identifiers to tracks with over 500,000 unique tags. Many of these tags are mood related, and more than 14,000 English words of these tags are used in terms of negative or positive features of the word, and two measures depending on the restlessness and energy to form the system It was scored with.
The Million Song database contains only the metadata of the song, not the song itself. The team linked all of this information to the Deezer catalog using identifiers such as song title, artist name, album title etc. Approximately 60% (18,644 tracks) of the resulting data set was used to train the AI and the rest were used to verify and further test the system.
Ultimately, researchers better detect whether the song is quiet or active rather than being able to detect whether the song is positive or negative, as well as conventional approaches that do not use AI I conclude that. . "This price rise seems to be due to the ability of our model to reveal and use the intermediary level correlation of voice and voice, especially when predicting valence.
Hope you like the news:
Deezer researchers have developed an AI system to detect musical emotions of songs
#Stay Tuned For More Updates :)