Artificial Intelligence Agents in Music Analysis: An Integrative Perspective Based on Two Use Cases
Antonio Manuel Martínez Heredia, Dolores Godrid Rodríguez, Andrés Ortiz García · 2025
Read Artificial Intelligence Agents in Music Analysis: An Integrative Perspective Based on Two Use Cases by Antonio Manuel Martínez Heredia, Dolores Godrid Rodríguez, Andrés Ortiz García (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-13987.txt.
Introduction
Antonio Manuel Martínez-Heredia, Dolores Godrid Rodríguez, and Andrés Ortiz García review AI agents for music analysis and test them in teaching settings. Their survey traces the field from rule-based systems through deep learning, multi-agent designs, and retrieval-augmented generation. They then evaluate educational uses through two concrete cases.
One case brings generative AI platforms into secondary classrooms to build analytical and creative skills. The other designs a multi-agent system for symbolic music analysis with modular, explainable workflows. Results show the agents improve pattern recognition, compositional parameterization, and feedback over older automated methods. The authors also flag open problems: transparency, cultural bias, and fair ways to judge hybrid human-machine work.
Catalog record
| CREATOR | Antonio Manuel Martínez Heredia, Dolores Godrid Rodríguez, Andrés Ortiz García |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-13987
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Artificial Intelligence Agents in Music Analysis: An Integrative Perspective Based on Two Use Cases Antonio Manuel Martínez Heredia, Dolores Godrid Rodríguez, Andrés Ortiz García Abstract This paper presents an integrative review and experimental validation of artificial intelligence (AI) agents applied to music analysis and education. We synthesize the historical evolution from rule-based models to contemporary approaches involving deep learning, multi-agent architectures, and retrieval-augmented generation (RAG) frameworks. The pedagogical implications are evaluated through a dual-case methodology: (1) the use of generative AI platforms in secondary education to foster analytical and creative skills; (2) the design of a multiagent system for symbolic music analysis, enabling modular, scalable, and explainable workflows. Experimental results demonstrate that AI agents effectively enhance musical pattern recognition, compositional parameterization, and educational feedback, outperforming traditional automated methods in terms of interpretability and adaptability. The findings highlight key challenges concerning transparency, cultural bias, and the definition of hybrid evaluation metrics, emphasizing the need for responsible deployment of AI in educational environments. This research contributes to a unified framework that bridges technical, pedagogical, and ethical considerations, offering evidence-based guidance for the design and application of intelligent agents in computational musicology and music education. Keywords : artificial intelligence, music analysis, education, deep learning, multi-agent systems 1 Introduction Despite considerable technological innovation, comprehensive reviews synthesizing the application and evolution of artificial intelligence (AI) in the field of music analysis remain scarce. Although early studies on computer-assisted composition and rule-based analysis established a foundation for the automated exploration of musical form and content 6 , there is still a limited body of literature addressing the complete progression from traditional algorithms to recent AI-driven models and hybrid systems. Pioneering work such as Miranda’s 16 , underscores the influence of AI, supercomputing, and evolutionary computation in shaping the first computational tools for creation. Recent reviews ( 20 ; 13 ) focus on intelligent music generation systems. However, a systematic integration of these historical advances with state-of-the-art AI methodologies and musical analysis is largely absent. In the last decade, deep learning frameworks—including convolutional neural networks, recurrent neural networks, and transformer architectures—have led to breakthroughs in music information retrieval. Recent advances also encompass interactive harmony tutoring ( 19 ) , structural music analysis ( 15 ) , intelligent education systems ( 5 ) , and the use of Large Language Model (LLM)-powered teachable agents in music pedagogy ( 9 ) . However, few studies provide an overarching review that brings together these disparate lines of inquiry or highlights the shift from task-specific solutions to holistic, explainable, and context-aware analysis. More recently, retrieval-augmented generation (RAG) models have introduced new standards of context awareness and interpretability by combining generative modeling with adaptive information retrieval. However, the literature lacks syntheses that discuss the joint impact of RAG with earlier and complementary AI-based approaches. A particularly groundbreaking development is the emergence of AI agents: autonomous systems that analyze, generate, and interact with music or learners using symbolic and audio data. The present review seeks to address this gap by providing a unified synthesis of the evolution of AI in music, from its historical roots to current advances in deep learning, agent-based systems, and RAG methodologies. We aim to clarify methodological innovations, highlight key research, and evaluate the implications for music research, analysis, and education going forward. The principal themes are the following: • The evolution of AI methods in music analysis. • The principles and applications of RAG methodologies. • Recent progress in deep learning and agent-based systems. • Implications for musicological research and pedagogy. This review aims to inform researchers, educators, and practitioners of current trends and emerging directions in computational musicology, thus filling a critical gap in the current academic landscape. This paper is organized as follows: Section I (Introduction) outlines the historical context and motivation to synthesize AI developments in music analysis. Section II traces the evolution of AI methods. Section III explains the methodology followed in this study. Section IV highlights two cases: the first is an exercise conducted with secondary students, and the second is an architecture designed to perform analysis. This section also details case studies and experimental results. Section V (Conclusions) summarizes key information and future research directions. The novelty of this review lies in the bridge between historical, technical, and pedagogical aspects of AI in music analysis through dual empirical case studies. 2 Related work In the field of music analysis, the evaluation of different types of music poses a significant challenge. One approach involves subjective evaluation, wherein individuals analyze music according to their own criteria. Another avenue involves establishing metrics and applying them systematically. Achieving the right balance between human feedback and quantitative indicators is essential. Human feedback provides invaluable insight into expressive and aesthetic aspects, often capturing nuances that computational methods may overlook. In contrast, quantitative indicators, such as accuracy, precision, recall, or diversity metrics, ensure reproducibility and objectivity when comparing systems. Integrating both perspectives enables comprehensive assessments and supports the development of robust, user-oriented music analysis tools. One of the main challenges in the objective evaluation of music analysis systems lies in the absence of a single universally accepted criterion to evaluate their performance. This study focuses on Western music analysis to provide a more unified approach for studying, appreciating, and understanding music. The characteristics of symbolic generation, structure, artistic expression, and aesthetics must therefore be examined through automatic methods. The computational evaluation of these characteristics has been revolutionized by recent advances in artificial intelligence. Deep learning architectures now represent the main generation methods for music composition systems. Enhancing the interpretability and controllability of deep networks is one of the future research directions for music generation technology 20 , which is particularly relevant given the need for objective but musically meaningful evaluation metrics discussed earlier. Multiple studies provide converging evidence that language and music share a cognitive system for structural integration, although the exact nature of this shared system remains complex and nuanced. Experimental evidence includes ( 3 ) , who found an interaction between linguistic and musical complexity in which comprehension accuracy changed when linguistic structures were paired with musically unexpected elements. 2.1 The evolution of AI methods in music analysis Early computational approaches to music analysis relied heavily on rule-based systems and expert heuristics, notably exemplified in Hiller and Isaacson’s computer-generated music experiments ( 6 ) and the generative theory of tonal music ( 14 ) . With the advent of the Music Information Retrieval (MIR) field—a multidisciplinary research area focused on developing innovative methods to analyze, search, and organize digital music data ( 1 ) —the focus shifted to designing features and algorithms—ranging from k-nearest neighbors to support vector machines—that could automatically classify, segment, or describe large corpora of music recordings and symbolic data. The subsequent explosion of large-scale annotated datasets—such as Musical Instrument Digital Interface (MIDI)-based datasets, the MIDI and Audio Edited for Synchronous TRacks and Organization (MAESTRO) ( 23 ) , the Guitar-Aligned Performance Scores (GASP) ( 18 ) , and MusicNet—and the increasing power of neural architectures ultimately enabled more flexible and data-driven models. Deep learning paradigms, by capturing complex temporal and harmonic structures in audio and scores, have largely superseded earlier approaches, although concerns around explainability and bias persist. Hybrid models that combine symbolic