aston89/UNlossifier-lossy-audio-reconstructor-and-sound-signature-simulator

UNlossifier is a U-Net based audio system for lossy reconstruction and sound signature simulation. It restores degraded compressed audio while also learning codec and analog-style transformations, enabling both audio recovery and creative sound domain transfer.

1 best points · 0 comments · 1 Hacker News post · 6 GitHub stars · View on GitHub

From the README

UNlossifier UNlossifier is an AI powered U-Net based audio system for lossy restoration and learned sound-domain transformation. Table of Contents * 1. Overview * 2. Key Features * 3. How It Works * 3b. Experimental Sound Signature Simulation: Limitations * 4. Audio Restoration Philosophy * 5. Model Architecture * 6. Training * 7. Custom Training (Core Feature) * 8. Usage * 9. Examples / Demos * 10. Limitations * 11. Potential * 12. Installation * 13. Usage Notes * 14. Safetensor Weights * 15. Notes * 16. Updates 1. Overview UNlossifier is an AI-driven audio restoration tool designed primarily to reconstruct high-quality sound from heavily compressed lossy files.

Hacker News discussions

• A neural network for lossy audio codec reconstruction (1 points, 0 comments, by Aston89, 2026-04-23)

More from aston89 on Hacker News

• aston89/CALP-Content-Aware-Loudness-Processor (2 points): A content-aware loudness processor with emergent non-linear behavior
• aston89/ASIP-Adaptive-Spatial-Imaging-Processor (1 points): Additive-only stereo imaging driven by time-frequency opportunity mapping
• aston89/FOE-compression-archetype (1 points): Foe: Functional Opcode Encoding