Thesis work - Automotive Audio Analysis using Music Information Retrieval

About this opportunity - Background

At Volvo Cars, delivering a premium in-vehicle audio experience is core to our brand identity. Historically, evaluating sound quality has relied on subjective listening tests—a process that is time-consuming, costly, and difficult to standardize. To innovate beyond these limitations, we are seeking to pioneer a data-driven approach that connects objective acoustic measurements with human perception.
The key to this innovation lies in Music Information Retrieval (MIR), the interdisciplinary science of extracting meaningful information from audio. While developed for tasks like genre classification, the power of MIR is its rich set of features that describe the character of a sound. The conceptual leap of this project is to treat our acoustic measurement data not as a simple graph, but as a "sound object" to be computationally described and categorized, just as MIR does with music. 

This project aims to bridge the gap between engineering and perception by creating a quantitative language for subjective descriptors. Ambiguous listener terms like "bright," "muddy," "harsh," or "punchy" can be systematically mapped to specific combinations of objective features (e.g., linking "brightness" to spectral centroid). This will create a common, precise language that connects our R&D teams with product marketing and, ultimately, our customers.

This thesis project is at the intersection of audio signal processing, data science, and automotive engineering. The core project is to explore, adapt, and apply MIR feature extraction methods, possibly including features from other fields, directly to our database of numerical acoustic measurements. This is a novel application, as we are not analysing full songs. Instead, the focus is on characterizing the sound system or Device Under Test (DUT) by analysing its detailed impulse response and transfer function.

The primary goal is to identify and engineer a set of robust features that can describe, classify and cluster our vehicle audio profiles. This will form the foundation for a scalable framework to:
•    Benchmark our sound systems with objective, repeatable metrics.
•    Identify and define a unique, data-driven "brand sound signature."
•    Implement a prototype database and dashboard for visualizing acoustic fingerprints and enabling interactive, complex queries (e.g., "What is the most bass-heavy sound system we have evaluated?").
•    Explore the reliability of automatic subjective tagging as a precursor to a future predictive model. Example tags could be dull, spatial, honky, punchy or boomy.

Scope of the thesis work

The student will work with our database of in-car acoustic measurements to:

•    Transform and Engineer Features: Convert raw numerical measurement data into a rich set of discrete MIR features (spectral, temporal, cepstral).
•    Apply Data Mining Techniques: Use clustering and classification algorithms to find similarities between car audio systems, identify natural groupings, and define acoustic profiles. 
•    Design and Implement Database & Dashboard: Structure the extracted features in a queryable database. Develop an interactive web-based dashboard to visualize the data, clusters, and allow for comparative analysis.
•    Investigate Temporal Dynamics: Use temporal modelling from MIR and other fields to analyse how sound quality characteristics might change under different conditions.

The project is structured into four main phases:
1.    Phase 1: Data Understanding and Preprocessing: Familiarization with the measurement database and development of a robust data cleaning and normalization pipeline.
2.    Phase 2: Feature Engineering and Extraction: Extraction of a comprehensive set of MIR features (e.g., MFCCs, spectral centroid, roll-off, flux) and engineering of novel features adapted for our specific signals.
3.    Phase 3: Modelling, Analysis, and Implementation: Application of unsupervised machine learning models (e.g., K-Means, DBSCAN, PCA, t-SNE) to cluster and analyse the feature data. Implementation of the database schema and development of the interactive dashboard prototype.
4.    Phase 4: Validation, Interpretation, and Thesis Writing: Correlation of findings with subjective data, interpretation of results from the dashboard and models, and final thesis composition.

Deliverables
•    A validated set of adapted features specifically tailored for in-car acoustic measurements.
•    A documented database schema for storing the extracted MIR features and associated metadata.
•    An interactive prototype dashboard for visualizing, comparing, and querying the acoustic characteristics of the audio systems.
•    A final Master's Thesis report and presentation of the findings to our engineering team.

What you'll bring

We are seeking a student pursuing a Master's degree in Electrical Engineering, Computer Science, Acoustics, Data Science, or a related field. The ideal candidate is passionate about audio technology and possesses a strong analytical mindset.

Required Skills:
•    Signal Processing: Strong theoretical and practical understanding of digital signals.
•    Data Science: Proficiency in machine learning, feature engineering, and data visualization.

Nice to have:
•    Python: Advanced programming skills with scientific computing and audio analysis libraries.
•    Databases / SQL: Experience in designing schemas and querying databases.
•    Dashboarding: Experience with creating interactive dashboards in Power BI.
•    Audio: Foundational knowledge of acoustics and psychoacoustics.

Duration

•    Tentative proposed thesis work period: 19th January 2026 to 26th of June 2026 (dates can be flexible with +/- 7 days)
•    Academic credits: equivalent to 30 ECTS 
•    Number of students: 1 student per project