Creating Effective Proposal Presentations 📊

This presentation introduces the practical design of a proposal focusing on Adaptive Gated Fusion within MoMuSE to address multimodal limitations. It highlights challenges with unreliable visual inputs affecting speech extraction. The solution integrates AGFN enabling context-based fusion, improving adaptability. The project explores enhancements in real-world applications, utilising adjusted encoding with datasets like VoxCeleb2. Success metrics include SI-SNR, SDR, PESQ, and STOI. Scheduled...