Music Dissertation Topics

Music Dissertation Topics

Info: Music Dissertation Topics
Published: 20th December in Music Dissertation Topics

Share this:

Facebook
Twitter
Whatsapp
Linkedin

Related Services

Our academic writing and marking services can help you

Music Dissertation Topics I phdassistance.com

Introduction

The progress made in the fields of music-related artificial intelligence, mental health research, and nursing education has been quite remarkable, but there are still wide gaps that prevent the use of technology and the advancement of education. To be precise, the latest research by Carone et al. (2025), Ching and Widmer (2025), and Ramírez-Moreno et al. (2025) mention challenges in the areas of abstract musical reasoning in audio LLMs, genre-biased music emotion recognition datasets, and the unreliable incorporation of mental health competencies in nursing curricula. These gaps highlight the urgent need for developing more solid frameworks, cross-genre generalisation methodologies, and cross-disciplinary collaborations that will not only enhance the reliability of models, consistency of education, and wider applicability across different cultures and contexts.

Dissertation Topic 1:

Developing Abstract Relational Reasoning Frameworks for Multimodal Audio Large Language Models in Music Perception

Background Context

The current technology has laid down a strong foundation and made remarkable strides in the area of audio Large Language Models research, yet they still struggle considerably in the domain of abstract musical reasoning tasks. In their study, The Muse Benchmark: Probing Music Perception in AI and Auditory Relational Reasoning in Audio LLMs (arXiv preprint arXiv:2510.19055), Carone, Roman, and Ripollés (n.d.) demonstrate that even the state-of-the-art models such as Qwen and Audio Flamingo 3 are incapable of executing the tasks of pitch-invariant melody recognition, harmonic function understanding, and rhythm synchronisation. Classic prompting techniques like Chain-of-Thought (CoT) and few-shot in-context learning often produce inconsistent or even counterproductive results, thus underlining a huge gap between machine and human performance. This points toward the need for a complete overhaul in model architecture and training approaches that will allow the understanding of deep auditory reasoning rather than relying on mere audio processing at the surface level.

PhD-Level Verification

It’s a fact that no current system can easily bring in abstract relational reasoning, along with the use of audio LLMs for the perception of intricate music. The study that we propose will not only evaluate new model structures but also provide novel training techniques aimed at narrowing the gap between humans and machines in the auditory perception area.

Research Questions
  • In what aspects, or rather, how can abstract relational reasoning be done by audio LLMs that work through multiple modalities in music recognition tasks?
  • What kinds of music representations can be said to be invariant, so much so that they would be considered as the model’s requirement to be able to generalise thoroughly across pitch, rhythm and harmony variations?
  • How do changes in architecture and the introduction of new training methods affect the models’ capability to reach the same performance level as humans in complex auditory reasoning tasks?
  • PhD-Level Contributions
  • The innovative development of audio LLM architectures or training methods suitable for performing abstract relational reasoning.
  • The assembly of benchmark datasets that will be useful for the evaluation of invariant musical perception in AI models.
  • The model performance, which was evaluated systematically, was measured against expert human perception in advanced music tasks.
    • Suggested Readings:

    Carone, B. J., Roman, I. R., & Ripollés, P. (2025). The muse benchmark: Probing music perception and auditory relational reasoning in audio llms. arXiv preprint arXiv:2510.19055.

    Dissertation Topic 2:

    Towards Invariant Musical Representation Learning in Audio Large Language Models for Advanced Music Understanding

    Background Context

    The current audio language models for music have very serious shortcomings in acquiring invariant representations, which are the basis for profound comprehension of music. Models such as Qwen and Audio Flamingo 3 are mentioned by Carone, Roman, and Ripollés (n.d.) in their paper The Muse Benchmark: Probing Music Perception and Auditory Relational Reasoning in Audio LLMs (arXiv preprint arXiv:2510.19055) as having abilities very close to random guessing in tasks like chord sequence matching, key modulation detection, and rhythm comparison. The prompting techniques of Chain-of-Thought and few-shot learning do, however, cut down the gap to some extent, but not to the level of human performance, thus making the perceptual defects clear. These results indicate the necessity of further work on invariant representation learning and architecture changes that would support the AI to be musically reasoning.

    PhD-Level Verification

    A universal method to integrate invariant musical representation learning into audio LLMs is still not available. The present study will investigate techniques for encoding invariances in pitch, rhythm and harmony to enhance AI’s human-like musical reasoning.

    Research Questions
  • Which is the best way for audio LLMs to learn invariant musical representations?
  • What transformations in model architecture or training would yield the most favourable conditions for abstract musical reasoning?
  • How much impact does the application of invariant representations have on model performance in challenging music tasks like chord recognition, melody perception, and key modulation detection?

    • PhD-Level Contributions

  • The creativity of new methods for the invariant audio LLMs musical representation learning.
  • Assessment of the model’s efficiency in complicated music perception and reasoning tasks.
  • A partnership in the creation of the next generation of AI-based music systems that will have human-like auditory reasoning capabilities.

    • Suggested Readings

    Carone, B. J., Roman, I. R., & Ripollés, P. (2025). The muse benchmark: Probing music perception and auditory relational reasoning in audio llms. arXiv preprint arXiv:2510.19055.

    Dissertation Topic 3:

    Closing the Data Distribution Gap in Music Emotion Recognition: Developing a Cross-Genre and Cross-Dataset Model for Universal Emotion Understanding

    Background Context

    The Music Emotion Recognition System has a very great future ahead, for it has already taken a number of unprecedented steps in the last few years. Still, limitations caused by the use of genre-specific datasets, differences in standards for annotation and inconsistency in the representation of emotions are among the major reasons MER is still not getting well in some areas. The study by Ching and Widmer (2025) in A Study on the Data Distribution Gap in Music Emotion Recognition (arXiv) reveals that the currently available MER models do not generalise well when tested against a dataset containing different musical styles. The problem is compounded by the use of separately processed valence-arousal labels and the dominance of certain genres (Pop, Jazz, Electronic) that cause distortions in emotional patterns, making it difficult for the models to generalise. Moreover, the voice quality, i.e., timbre, is one of the factors that affect the emotions of the listener, but this is not represented symbolically, and the use of outdated feature analysis tools is one of the limitations to reproducibility. The aforementioned limitations of the MER technology, therefore, indicate a need for a genre-inclusive, unified approach to music emotion recognition that is not constrained by genres.

    PhD Level Verification

    Currently, there is no comprehensive method that deals with the distribution divergence between different MER data sets. Moreover, no method allows generalised prediction of emotion across types. Such a model design that is free of genre bias and inconsistency in annotation is a task for a PhD student.

    Research Questions
  • What could be a systematic way of eliminating cross-genre emotion recognition and cross-dataset inconsistencies in emotion annotations in MER?
  • Which modelling techniques can facilitate generalizability over a broad range of musical styles and audio qualities that are expressive?
  • Is it possible to create a unified musical emotion that stays valid throughout different culturally and stylistically diverse datasets?
    • PhD-Level Contributions
  • An all-inclusive MER framework was developed that unifies the emotional annotations that are present in different datasets.
  • A cross-genre training strategy was developed to diminish the bias caused by the different data distributions.
  • New cross-dataset benchmarks helped in the validation of the generalizable MER models.
    • Suggested Readings

    Ching, J., & Widmer, G. (n.d.). A study on the data distribution gap in music emotion recognition. Institute of Computational Perception, Johannes Kepler University Linz; LIT AI Lab, Linz Institute of Technology. https://arxiv.org/abs/2510.04688

    Dissertation Topic 4:

    Designing Genre-Invariant Emotion Annotation Frameworks: Reducing Bias and Improving Consistency in Music Emotion Recognition Systems

    Background Context

    The research in the Music emotion annotation framework are very different from each other across datasets and is affected by the culture, style, and perception of the annotators. Ching and Widmer (2025) explain that using inconsistent valence–arousal scales and applying different normalisation methods independently lead to emotion distributions that are incorrectly seen as similar, although they are very different in content. The predominance of certain genres and the subjectivity in annotations add to the confusion regarding the emotions present in MER datasets and, consequently, the impossibility of achieving a universal or genre-invariant emotion recognition. Models that rely on audio exacerbate this problem due to the strong emotional impact of timbre, whereas symbolic data only conveys a limited range of expressive features. The discipline is short of such annotation systems that are standard and reliable and can work across different musical traditions.

    PhD-Level Verification

    PhD-Level Verification: A systematic appraisal of the impact of behavioural and environmental interventions, evaluation of technological tools and an integrated model for mental health protection in populations exposed to noise are required for a PhD project. This requires collaboration between different disciplines such as psychology, environmental health, neurophysiology, and acoustic technology.

    Research Questions
  • What are the most effective behavioural interventions (e.g., exercise, meditation, resilience training) that can help to prevent mental health problems caused by chronic noise exposure?
  • In what way do psychological factors that are moderated by environmental factors, for instance, access to green spaces, affect noise?
  • What are the benefits of noise reduction technologies for mental well-being that can be measured in everyday situations?
    • Contributions at the PhD-Level
  • A holistic multi-variable model for the evaluation of reducing actions from the three areas of behaviour, environment, and technology.
  • Research findings on the shielding mental health impact of physical workouts, mindfulness practice, and resilience-focused interventions.
  • Assessment of the technologies for diminishing noise and their benefits for mental health.
    • Suggested Readings

    Ching, J., & Widmer, G. (n.d.). A study on the data distribution gap in music emotion recognition. Institute of Computational Perception, Johannes Kepler University Linz; LIT AI Lab, Linz Institute of Technology.  https://arxiv.org/abs/2510.04688

    Dissertation Topic 5:

    Towards a Standardised, Perception-Aligned, and Cross-Cultural Framework for Evaluating Music Generation Systems

    Background Context

    AI-based music generation evaluation methods have shown considerable progress within a short time, yet the evaluation methods still differ a lot, and their reliability is questionable. Kader and Karmaker (n.d.) comment that there is no accepted universal framework that can cover the most important musical qualities like structure, creativity, emotional expression, or coherence. The use of existing objective metrics is often the case of very poor correlation with human perception, and besides that, they are subjected to Western-centric biases that make it difficult to evaluate genres from non-Western and low-resource areas. Also, symbolic music evaluation still spins in its own little world while audio-based approaches are far more advanced; it adopts very basic measures that disregard the higher-level musical aspects. Filling these gaps requires a perception-aligned, culturally inclusive evaluation framework to be developed.

    PhD-Level Verification

    The future research plans the creation and validation of a wide-ranging evaluation framework, which will incorporate computation metrics, human listener experiments and cultural standards for symbolic and audio music generation. The main emphasis in the doctoral contribution is on the issues of normalisation, perceptual legitimacy and cultural diversity.

    Research Questions
  • What methods could be developed for creating objective evaluation metrics that would be more in accordance with the perception of quality, creativity, and emotional expression in music by humans?
  • Which framework can help in creating a standardised and comparable evaluation across different music generation models and tasks?
  • In what way do the evaluation metrics differ in their performance for Western and non-Western genres of music, and what are the effective ways of minimising cross-cultural bias?
  • What are the key areas of improvement that need to be targeted so that symbolic music evaluation would attain a level of sophistication equivalent to that of audio-based methods?
    • PhD-Level Contributions
  • A common, perception-aligned standardised metric for music generation systems.
  • Research findings that provide a direct link between computational metrics and human listener judgments.
  • Benchmark tests for cross-cultural evaluation that directly deal with the Western bias in music generation research.
  • Advanced techniques for symbolic music evaluation that integrate the higher-level musical features.
    • Suggested Readings

    Kader, F. B., & Karmaker, S. (n.d.). A survey on evaluation metrics for music generation. https://arxiv.org/abs/2509.00051

    Conclusion

    Need assistance finalising your dissertation topic? Selecting a strong, researchable topic can be challenging — but you don’t have to do it alone.
    Our research consultants can help refine your ideas, identify literature gaps, and guide you toward a topic that aligns with current academic trends and your programme requirements.
    Contact us to begin one-on-one topic development and refinement with PhdAssistance.com Research Lab.

    Share this:

    Facebook
    Twitter
    Whatsapp
    Linkedin

    Cite this work

    Study Resources

    Free resources to assist you with your university studies!

    Research Questions