Chaos-Inspired AI Framework for Audio Bandwidth Extension
A novel adversarial framework known as NDSI-BWE has been developed for bandwidth extension (BWE), aimed at restoring high-frequency elements that bandwidth limitations have obscured. This framework utilizes four innovative discriminators derived from nonlinear dynamical systems: the Multi-Resolution Lyapunov Discriminator (MRLD), which focuses on sensitivity to initial conditions through deterministic chaos; the Multi-Scale Recurrence Discriminator (MS-RD) for self-similar recurrence dynamics; the Multi-Scale Detrended Fractal Analysis Discriminator (MSDFA) for long-range scale-invariant relationships; and the Multi-Resolution Poincaré Plot Discriminator (MR-PPD) for uncovering hidden latent space connections. Furthermore, it features a Multi-Period Discriminator (MPD) to identify cyclical patterns and a Multi-Resolution Amplitude Discriminator (MRAD). This research, applicable in fields from telecommunications to high-fidelity audio on constrained resources, is available on arXiv with the identifier 2507.15970.
Key facts
- NDSI-BWE is a new adversarial bandwidth extension framework.
- It uses four chaos-inspired discriminators: MRLD, MS-RD, MSDFA, MR-PPD.
- Also includes MPD and MRAD discriminators.
- Aims to recover high-frequency audio components lost due to bandwidth constraints.
- Applications include telecommunications and high-fidelity audio.
- Published on arXiv with ID 2507.15970.
- The discriminators are based on nonlinear dynamical systems.
- MRLD captures deterministic chaos via Lyapunov exponents.
Entities
Institutions
- arXiv