This paper provides a systematic exploration of Artificial Intelligence (AI) for music generation within a cloud-edge synergy architecture (i.e. a coordinated cloud–edge computing paradigm where the cloud handles compute-intensive training and high-fidelity generation, while the edge supports latency-sensitive and privacy-critical inference and interaction), focusing on its deep integration and practical applications across educational and performance ecosystems. It begins by reviewing the evolution of key technologies, from early Recurrent Neural Networks (RNNs/LSTMs) and Generative Adversarial Networks (GANs) to the current mainstream Transformer architectures and Diffusion Models, highlighting breakthroughs in handling long-range dependencies, fidelity, and controllability. The core of the paper delves into cloud-edge collaborative system architectures designed for scenarios involving music generation, real-time interaction, and multimodal alignment. It provides an in-depth analysis of four primary collaboration models—Edge-first, Cloud-first, Split Inference/Learning, and Federated Personalization—detailing their characteristics, suitable application scenarios, and inherent trade-offs. Furthermore, the paper comprehensively examines how this architecture empowers the educational ecosystem by enabling resource democratization, personalized learning, and pedagogical innovation. Finally, the paper identifies prevailing challenges at the system, model, ethical/copyright, and privacy/security levels, and outlines future research directions, including model-system co-design, multi-modal and XR integration, trustworthy copyright mechanisms, and standardized open-source ecosystems. This work argues that cloud-edge synergy is the key enabling technology for the scalable and deep application of AI-driven music creation in latency-sensitive and privacy-critical domains such as education and performance.
Fan et al. (Mon,) studied this question.