Deezer develops AI detection tool for explicit song lyrics

The system uses machine learning to detect explicit lyrics through audio, and no additional metadata.

Tagging songs as “explicit” sometimes requires a high-level understanding of cultural expectations and could involve a lot of subjectivity. With the service, the music streaming platform wanted to create technology that assists the person responsible with this process.

To do this Deezer developed a tool called Spleeter, which is used to extract vocals from songs. It’s freely available for anyone to use and is reasonably competent. After the extraction, the lyrics are run through the keyword spotting system that flags potentially offensive words. Then a binary classifier determines whether the words are explicit or not.

This is the first attempt to create an explicit content detector purely based on audio and Deezer says it hopes the new method of identifying explicit lyrics will reduce bias and improve accuracy.

Removing bias is a common challenge in machine learning. A model can adapt to biases present in sample data, which is a phenomenon called “horse” in music analysis. This could cause bias towards a particular music genre.

“If you’re not careful when designing your experiment, you may end up with a system that instead of detecting explicit lyrics, will detect rap songs, just because they have on average, a higher probability of containing explicit lyrics,” Deezer director of research Manuel Moussallam said.

Although the technology is not yet "fit for tagging songs as explicit in a fully automated manner," it could be used to help humans identify potentially explicit material.

Deezer's full research paper will be published at the International Conference on Acoustics, Speech, and Signal Processing, which will be held virtually until 8 May.