Master's Thesis: Differentiating Music via Deep Representation Learning

What characterizes a musical work? In our music catalog of over 100 million tracks, a single musical work like "Bohemian Rhapsody" by Queen can exist in dozens of versions: the original studio recording, various remasters, a live performance, and even covers by other artists. Managing this complexity is a deep and fascinating challenge for music streaming companies such as Soundtrack.

Traditional audio fingerprinting is excellent at identifying identical recordings, but it fails to understand that a studio original and a live acoustic version are, in essence, the same musical work. Metadata can help, but it can also be inaccurate or misleading. This limitation leads to a fragmented catalog, which negatively impacts everything from search functionality to user recommendations. The challenge is to create an intelligent system that can operate on both levels: recognizing the fundamental musical "work" while also distinguishing the unique characteristics of each specific "recording."

Solving this will have a massive impact, leading to:

• A more organized and cleaner music catalog.
• Vastly improved music discovery and recommendation systems.
• A richer, more intuitive browsing experience for our users.

This Master's Thesis project centers on Deep Representation Learning to distinguish between a musical "work" and a specific "recording." Your primary task is to develop a model that generates hierarchical audio representations. At one level, these representations will group all versions of a song—from studio originals to live covers—as the same fundamental "work." At a more detailed level, they will differentiate the unique sonic signatures of each individual "recording." You will leverage state-of-the-art deep learning architectures (e.g., Conformers) to solve this complex problem and contribute a powerful new tool for music catalog management. As a member of our Content Services team, you will collaborate with machine learning engineers and music experts to apply your research to a real-world challenge.

You're driven and entrepreneurial, but you know how to be a team player too. Just like us, you love music. Regardless of roles, we're always looking to work with people who can adapt to constant change, prioritise what's important, stay humble, open, curious and with a passion for details.

This opportunity is for you if you have:

• A strong passion for machine learning and a curiosity for its theoretical foundations.
• Experience coding in Python and using deep learning frameworks (e.g., PyTorch, TensorFlow).
• An interest in representation learning, self-supervised methods, or audio processing.
• The ability to communicate complex research topics clearly and effectively.
• Fluency in English.
• Residency or citizenship in Sweden, preferably near Stockholm.

Soundtrack is a B2B scale-up company providing music streaming services for businesses. We serve small customers like the café around the corner, and much bigger brands like Joe & the Juice, Toni & Guy and TAG Heuer. On the inside, we're a bunch of talented, motivated and humble designers, engineers and music experts. We believe in product-led growth, where the product is the primary driver of customer acquisition, conversion and expansion.

The Content Services team manages our music catalog—the backbone of our service. They are responsible for ingesting, managing, and delivering our library of over 100 million tracks. The robust and accurate music data they provide is essential to our entire service, powering everything from search to our recommendation engines.

A Master's Thesis is an excellent way for us to get to know new talent. We believe that diversity of perspective and experience makes our team and our product better, and we encourage you to apply. At Soundtrack, we believe that diversity fuels creativity and innovation and are committed to building a team that reflects a variety of backgrounds, perspectives, and experiences.

This is a full-time thesis project intended for the Spring 2026 semester. If this sounds like the perfect final project of your master's degree, and a challenge you'd love to take on, we encourage you to apply. We reserve the right to close this vacancy early if we identify a suitable candidate before the application deadline. To ensure consideration, we encourage you to submit your application as soon as possible.

If you have any questions about the position or need to reach out, get in touch with Adam Skärbo Jonsson at Please note that we only accept applications submitted via our career page and do not accept applications by e-mail.