ReMi: A Random Recurrent Neural Network Approach to Music Production

2025-04-02

arXiv.org

AI Summary

Generative AI music synthesis faces challenges including high energy consumption, copyright infringement risks, and constrained creative expressivity. To address these, this work proposes a training-free, data-agnostic stochastic recurrent neural network (RNN) architecture—employing fully randomly initialized LSTM or GRU units—integrated with temporal parameterization control and real-time audio synthesis interfaces. The method enables low-latency (<10 ms), milliwatt-level power consumption for controllable musical signal generation, bypassing end-to-end learning paradigms entirely. It supports interactive, improvisational generation of configurable musical elements—including arpeggios and low-frequency oscillators (LFOs)—while drastically reducing computational and energy overhead and eliminating dataset-dependent copyright liabilities. The implementation is lightweight and has been integrated into an open-source music production workflow platform. This establishes a novel human-AI co-creation paradigm requiring zero model training and imposing no copyright burden on musicians.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Generative artificial intelligence raises concerns related to energy consumption, copyright infringement and creative atrophy. We show that randomly initialized recurrent neural networks can produce arpeggios and low-frequency oscillations that are rich and configurable. In contrast to end-to-end music generation that aims to replace musicians, our approach expands their creativity while requiring no data and much less computational power. More information can be found at: https://allendia.com/

Problem

• Addresses high energy consumption in generative AI music production
• Reduces copyright risks by avoiding data-dependent training
• Enhances musician creativity with minimal computational resources

Innovation

• Randomly initialized recurrent neural networks
• No data required for operation
• Low computational power consumption