[2603.29326] Real-Time Band-Grouped Vocal Denoising Using Sigmoid-Driven Ideal Ratio Masking

Computer Science > Sound arXiv:2603.29326 (cs) [Submitted on 31 Mar 2026] Title:Real-Time Band-Grouped Vocal Denoising Using Sigmoid-Driven Ideal Ratio Masking Authors:Daniel Williams View a PDF of the paper titled Real-Time Band-Grouped Vocal Denoising Using Sigmoid-Driven Ideal Ratio Masking, by Daniel Williams View PDF HTML (experimental) Abstract:Real-time, deep learning-based vocal denoising has seen significant progress over the past few years, demonstrating the capability of artificial intelligence in preserving the naturalness of the voice while increasing the signal-to-noise ratio (SNR). However, many deep learning approaches have high amounts of latency and require long frames of context, making them difficult to configure for live applications. To address these challenges, we propose a sigmoid-driven ideal ratio mask trained with a spectral loss to encourage an increased SNR and maximized perceptual quality of the voice. The proposed model uses a band-grouped encoder-decoder architecture with frequency attention and achieves a total latency of less than 10,ms, with PESQ-WB improvements of 0.21 on stationary noise and 0.12 on nonstationary noise. Subjects: Sound (cs.SD); Artificial Intelligence (cs.AI) ACM classes: I.2.m; I.5.1 Cite as: arXiv:2603.29326 [cs.SD]   (or arXiv:2603.29326v1 [cs.SD] for this version)   https://doi.org/10.48550/arXiv.2603.29326 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Daniel Willi...