Research map / Architectures and efficiency / RNNs and normalization
SFSRNet: Super-resolution for Single-Channel Audio Source Separation
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 |
|---|---|---|---|
| add SFSRNet with dynamic mixing | SepFormer + DM · WSJ0-2mix | SI-SNRi (dB): 22.3 -> 24.0 | Helped |
| add multi-loss with same sampling rates (ablation) | SepFormer + DM · WSJ0-2mix | SI-SNRi (dB): 22.3 -> 23.0 | Helped |
| use increasing sampling rates for multi-loss estimates (ablation) | SFSRNet multi-loss · WSJ0-2mix | SI-SNRi (dB): 23.0 -> 23.3 | Helped |
| add SR without heuristics (ablation) | SepFormer + DM · WSJ0-2mix | SI-SNRi (dB): 22.3 -> 22.9 | Helped |
| add heuristics to SR (ablation) | SFSRNet with SR · WSJ0-2mix | SI-SNRi (dB): 22.9 -> 23.3 | Helped |
| add multi-loss and heuristic-guided SR (ablation) | SepFormer + DM · WSJ0-2mix | SI-SNRi (dB): 22.3 -> 24.0 | Helped |
| downsample DPRNN with stride 8 instead of stride 1 | DPRNN · WSJ0-2mix | SI-SNRi (dB): 19.1 -> 18.0 | Hurt |
| add SR after stride-8 downsampling | DPRNN · WSJ0-2mix | SI-SNRi (dB): 18.0 -> 18.4 | Helped |
| reduce DPRNN chunk size from 200 to 50 | DPRNN, stride 8 · WSJ0-2mix | SI-SNRi (dB): 18.0 -> 17.7 | Hurt |
| add SR with chunk size 50 | DPRNN, stride 8 · WSJ0-2mix | SI-SNRi (dB): 17.7 -> 18.8 | Helped |
| add dynamic mixing | SFSRNet · clean Libri2Mix | SI-SNRi (dB): 20.4 -> 21.7 | Helped |
| add dynamic mixing | SFSRNet · noisy Libri2Mix | SI-SNRi (dB): 15.6 -> 16.4 | Helped |
Closest papers by meaning
- Audio Super-Resolution with Latent Bridge Models (NeurIPS 2025 · similarity 0.87)
- ZeroSep: Separate Anything in Audio with Zero Training (NeurIPS 2025 · similarity 0.84)
- Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation (NeurIPS 2024 · similarity 0.84)
- TIGER: Time-frequency Interleaved Gain Extraction and Reconstruction for Efficient Speech Separation (ICLR 2025 · similarity 0.83)
- SSLAM: Enhancing Self-Supervised Models with Audio Mixtures for Polyphonic Soundscapes (ICLR 2025 · similarity 0.83)
- ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior (ICML 2025 · similarity 0.83)
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.