Home
This vault documents PhD research in the CHAIS project (Crossmodal Human-AI Collaboration for Selective Hearing Assistance) at Universität Hamburg: neuro-inspired speech enhancement for smart glasses that help listeners focus on one speaker in noisy, multi-talker scenes (the Cocktail Party Problem). It contains the project documentation, the research on oscillatory state-space models for speech enhancement, and a general wiki on speech processing and machine learning.
For a quick overview:
- Research Summary for a one-page summary of the project, its research plan and the current state of the work.
- CHAIS Overview, then Cocktail Party Problem and Research Plan for the project goals, background and phases.
- Oscillatory SSM project for the current research, with background in Speech Enhancement, State-Space Models, SEMamba and LinOSS.
- Infrastructure for where code, experiment logs, machines and data live.
Table of Contents¶
Project¶
- Research Summary Consolidated summary of the CHAIS project (proposal background, objectives, work packages), its current research plan, and the oscillatory SSM architecture work within it.
- CHAIS Project Project overview with knowledge, methodology and organization notes, roadmap and milestones.
- Oscillatory SSM for Speech Enhancement Research question, results, theory and status of the oscillatory SSM work.
- Ideas Research ideas that have not grown into projects yet.
- Infrastructure Repositories, experiment tracking, machines, drives and datasets.
Knowledge Base¶
- SSMs for Speech Enhancement Review of the SSM-based speech enhancement frontier (2026).
- Wiki General knowledge base about signal processing, ML models and neuroscience, including State-Space Models and literature notes.