This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Neural-Enhanced Dynamic Range Compression Inversion: A Hybrid Approach for Restoring Audio Dynamics
Created by
Haebom
Author
Haoran Sun, Dominique Fourer, Hichem Maaref
Outline
This paper presents a hybrid approach combining model-based inverse transform and neural networks to solve the inverse transform problem of dynamic range compression (DRC). To overcome the limitations of existing methods, which overlook key parameters or rely on precise parameter values, we use neural networks (classification and regression) to estimate DRC parameters and reconstruct the original signal. Experimental results on various music and speech datasets demonstrate that the proposed method outperforms existing state-of-the-art techniques in terms of performance and robustness.
Takeaways, Limitations
•
Takeaways:
◦
We present a novel hybrid approach that combines model-based DRC inversion and neural networks to simultaneously perform DRC parameter estimation and audio restoration.
◦
Overcomes the dependence on exact parameter values, a limitation of existing methods, and achieves robust performance.
◦
Demonstrated superior performance to existing state-of-the-art techniques on various music and speech datasets.
◦
Contributes to restoring original dynamics, improving remixing, and enhancing overall audio quality in various fields such as music production, broadcasting, and voice processing.
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed neural network architecture.
◦
Applicability and performance evaluation of various DRC algorithms and settings is needed.
◦
Further research is needed to evaluate performance in real acoustic environments and to determine practical applicability.
