Neural Acoustic Feedback Cancellation: From Signal Processing to Deep Learning (video)

This presentation explores recent advancements in integrating traditional signal processing with deep learning to address the challenges of acoustic feedback cancellation, particularly focusing on acoustic echo and howling suppression. While traditional methods typically rely on adaptive filters to mitigate these issues, their performance is often limited by nonlinear distortions. With the rapid progress in deep learning, numerous studies have proposed solutions that either leverage deep learning independently or combine it with conventional adaptive filtering techniques. The presentation begins by outlining the fundamental problem of acoustic feedback cancellation, emphasizing its key technical challenges and the complexities involved in real-time processing. Following this, I will introduce several innovative hybrid approaches that integrate deep learning with traditional signal processing methods, delving into their underlying methodologies and evaluating their effectiveness in addressing feedback cancellation issues. Through theoretical analysis and experimental results, I will discuss the key challenges these methods face in tackling feedback cancellation, as well as their potential and advantages in real-world applications.