Reproduce “MC-SEMamba: A Simple Multi-channel Extension of SEMamba” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wen-Yuan Ting, Wenze Ren, Rong Chao, Hsin-Yi Lin, Yu Tsao, Fan-Gang Zeng
MC-SEMamba extends the Mamba-based SEMamba model to multi-channel speech enhancement by adapting its input layer to handle multiple audio streams. The model achieves competitive performance on the CHiME3 dataset, leveraging spatial information effectively while maintaining high parameter efficiency.
Key insights
- The Mamba architecture, known for its linear complexity, is well-suited for efficient multi-channel speech enhancement tasks.
- A straightforward adaptation, primarily modifying the input layer, allows a single-channel Mamba-based model to effectively process and leverage multi-channel spatial information.
- Performance improvements are observed with an increasing number of microphones, with results suggesting an optimal balance around five microphones for many applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.