Research map / Architectures and efficiency / CNN design
Speech Separation Using an Asynchronous Fully Recurrent Convolutional Neural Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| asynchronous (A-FRCNN) instead of synchronous updating (ablation) | FRCNN · Libri2Mix | SI-SNRi: 12.1 -> 15.5 | Helped |
| add bottleneck design at bottom stage (ablation) | S-FRCNN · Libri2Mix | SI-SNRi: 12.1 -> 14.6 | Helped |
| remove bottom-up skip-connections (ablation) | Control 1 · Libri2Mix | SI-SNRi: 14.6 -> 14.6 | No clear effect |
| add delayed global fusion (ablation) | Control 2 · Libri2Mix | SI-SNRi: 14.6 -> 15.5 | Helped |
| add delayed global fusion to U-Net (ablation) | U-Net · Libri2Mix | SI-SNRi: 11.4 -> 13.2 | Helped |
| reduce channels of S-FRCNN to match params (ablation) | S-FRCNN · Libri2Mix | SI-SNRi: 12.1 -> 11.8 | Hurt |
| remove skip connections from S-FRCNN (ablation) | S-FRCNN · Libri2Mix | SI-SNRi: 12.1 -> 11.43 | Hurt |
| number of stages S=5 vs 4 (ablation) | A-FRCNN · Libri2Mix | SI-SNRi: 14.4 -> 15.5 | Helped |
| number of stages S=6 vs 5 (ablation) | A-FRCNN · Libri2Mix | SI-SNRi: 15.5 -> 14.4 | Hurt |
| summation connection instead of direct connection (macro unfolding) (ablation) | A-FRCNN · Libri2Mix | SI-SNRi: 12.8 -> 15.5 | Helped |
| unfold 16 steps instead of 8 | A-FRCNN · Libri2Mix | SI-SNRi: 15.5 -> 16.7 | Helped |
| replace concatenation with summation in fusion (ablation) | A-FRCNN-16 · Libri2Mix | SI-SNRi: 16.7 -> 16.2 | Mixed |
Closest papers by meaning
- Knowing When to Quit: Probabilistic Early Exits for Speech Separation Networks (ICLR 2026 · similarity 0.86)
- Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation (NeurIPS 2024 · similarity 0.86)
- Voice Separation with an Unknown Number of Multiple Speakers (ICML 2020 · similarity 0.86)
- RTFS-Net: Recurrent Time-Frequency Modelling for Efficient Audio-Visual Speech Separation (ICLR 2024 · similarity 0.85)
- TIGER: Time-frequency Interleaved Gain Extraction and Reconstruction for Efficient Speech Separation (ICLR 2025 · similarity 0.84)
- Accelerating Linear Recurrent Neural Networks for the Edge with Unstructured Sparsity (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.