Research internship - AI music detection (m/f/d)

Recent advances in music generation have enabled the development of systems capable of producing very realistic music: numerous platforms now offer services that let you generate a song as an audio file from a simple prompt describing the intended results. While the technical progress is impressive, the cheap and massive generation of music has also downsides, and AI-generated music now floods music streaming platforms (Deezer recently reported that more than a third of its daily deliveries are AI-generated), containing mainly AI slop with little legitimate use or even fraudulent use.

To monitor this massive flow as well as to be transparent with the end user, Deezer developed a detector of AI-generated content able to detect content generated by various platforms with very high accuracy. Nevertheless, the system can be fooled by some audio manipulations [1] and may not be able to give further insights in cases of hybrid music tracks containing both AI-generated and recorded stems or samples.

The objective of the internship will be to study the detection of various cases of hybrid content in a controlled context.

This position has a start date between beginning of February and end of April 

What you will do:

• Collect data to generate a dataset of hybrid content

• Propose and discuss solutions to the task at hand.

• Develop and evaluate solutions.

[1] Darius Afchar, Gabriel Meseguer-Brocal, Romain Hennequin. AI-Generated Music Detection and Its Challenges. ICASSP 2025