Music Information Retrieval (MIR) is essential for comprehending and analysing music, and it has various applications in music education, music creation, music recommendation, and other related areas. Conventional music processing techniques heavily depend on human derived characteristics and regulations, which hinders the comprehensive exploration of the abundant information embedded in music. This thesis aims to utilise artificial intelligence approaches, specifically modelling methods rooted in music knowledge and cognition, to address three objectives: automatic music transcription, predominant instrument detection, and music shape evaluation. Automatic music transcription (AMT) is the process of effectively identifying notes from audio signals. Predominant musical instrument recognition (PMIR) involves determining the dominant instrument in a musical section. Music shape evaluation (MSE) shows performance qualities and styles. This thesis introduces a cognitionguided framework for AMT, achieving F-measures of 76.3% on the MAPS dataset (an 8% improvement over the baseline), 80.17% on the BACH10 dataset (second-best performance), and 67.63% on the TRIOS dataset (leading performance). For PMIR, an innovative HHT-DCNN framework is proposed, achieving an 84% F-measure on the IRMAS dataset, which represents a 6% improvement over state-of-the-art methods. Finally, a new dataset is created for the MSE task, and a novel S-ResNN architecture is introduced, achieving an average accuracy of 93.78% across different training ratios. The experimental findings indicate that the suggested approaches may greatly improve current technical standards and achieve outstanding performance. Moreover, the findings of this thesis have the potential to be applied in several aspects of music education, such as the creation of curriculum, the development of interactive learning tools, and the design of personalised music training programmes. This thesis focuses on computational music comprehension and offers substantial contributions to automatic music transcription, instrument recognition, and performance analysis. It highlights the importance and potential applications of research in computational musicology.
| Date of Award | 20 Feb 2025 |
|---|
| Original language | English |
|---|
| Awarding Institution | - University Of Strathclyde
|
|---|
| Sponsors | University of Strathclyde |
|---|
| Supervisor | Stephan Weiss (Supervisor) & John Soraghan (Supervisor) |
|---|