Recent advancements in machine learning-based speech enhancement focus on sophisticated deep learning models—including deep neural networks (DNNs), transformer architectures, and generative models—engineered for robust, real-time performance, personalization, and adaptability across varied acoustic environments.

The presentation will address key developments in speech enhancement such as:

• Quality-optimized and task-specific model architectures
• Real-time and edge-oriented implementations
• Multichannel and spatially aware signal processing
• Generative data augmentation methods
• Applications of self-supervised and foundation models
• Unified frameworks supporting both personalized and general speech enhancement