Shane Foster-Smith

1.

Direct sampling refers to the process of extracting a section of audio from another song and, usually after various manipulations, incorporating it into one or more places in a new song. This production technique is ubiquitous in modern music, and, when done right, can recontextualize/reimagine a piece of older music and credit/compensate the original artist whose music was sampled. However, often direct sampling is not done ethically. In cases where popular artists sample other popular artists unethically, these cases are usually resolved with the song being taken down from streaming platforms or clearance is later provided by artist or label that owns the sampled song’s master. However, if a bigger artist unethically samples a smaller artist’s work, what recourse does the smaller artist, who likely doesn’t have the resources to fight a drawn-out legal battle with a major label or artist, have? These cases should be, and usually are, handled by the streaming services. Theoretically, a smaller artist uploads a complaint to the streaming service and then, someone employed at the streaming service would manually confirm that the song sampled the original artist’s work without clearance and remove that song from their platform. The issue with this solution is that someone employed for the streaming service must manually confirm if another song samples another song. As we will in this project, detecting certain samples often requires some musical training and/or a good ear. These music streaming services simply don’t have enough trained personnel to handle each case correctly/timely.

2.

In this project, I propose an algorithm and implementation (in PowerPoint) that detects where in an original piece of music the sampling occurred and how that section was manipulated to be put into the new song. Later, I consider how an algorithm like this might be used to aid in music training and discovery, plagiarism/clearance detection for streaming platforms and for increasing the recognition and compensation of smaller artists, who often struggle in a business dominated by the bigger artists and record labels.

3.

The main challenge in this project was optimizing the mapping from the sampled song to the new song and considering supervised learning model for the similarity functionality. Audio analysis is conceptually, computationally and mathematically complex, so learning about all these different ideas was challenging!

4.

The next step for this project is to collect the necessary data to determine the best optimizations for the mapping function and for the supervised machine learning model. Luckily, there are thousands of manually made examples online, however it will take a significant amount of time (weeks) to collect the necessary information. After that, coding!

Built With

• librosa
• powerpoint
• python
• whosampled.com