Real-Time Band-Grouped Vocal Denoising Using Sigmoid-Driven Ideal Ratio Masking
This addresses latency issues in live vocal denoising applications, though it is incremental over existing deep learning methods.
The paper tackled real-time vocal denoising by proposing a sigmoid-driven ideal ratio mask with a band-grouped encoder-decoder architecture, achieving a latency under 10 ms and PESQ-WB improvements of 0.21 on stationary noise and 0.12 on nonstationary noise.
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.