Research interests

• Diffusion Models
• Speech Enhancement
• Multimodal Speech Processing
• Knowledge Localization in Deep Neural Networks

Publications & Patents

1.  Sina Khanagha, Timo Gerkmann, "Your U-Net Dereverberation Model is Secretly an RIR Encoder", ISCA Interspeech, Sydney, Australia, 2026. [arxiv]

2. Sina Khanagha, Bunlong Lay, Timo Gerkmann, "Bone-Conduction Guided Multimodal Speech Enhancement with Conditional Diffusion Models", ICASSP, Barcelona, Spain, 2026. [arxiv]

3.  Jakob Kienegger, Tal Peer, Sina Khanagha, Timo Gerkmann, "Weakly Guided and Autoregressive Beamformer Parameterization for Generalizable Moving Speaker Extraction in Higher-Order Ambisonics", IWAENC, Cremona, Italy, 2026. [arxiv]

4. Daniela Hall, Soufiane Atouani, Remi Poncot, Sina Khanagha, "Enhanced Sound-Related and Gesture-Related Authentication", U.S. Patent Application, US20260030332A1, 2026. [patent]