Research map / Architectures and efficiency / CNN design
Deep Residual-Dense Lattice Network for Speech Enhancement
Mohammad Nikzad, Aaron Nicolson, Yongsheng Gao, Jun Zhou, Kuldip K. Paliwal, Fanhua Shang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| use the RDL-Net rather than Metric-GAN | speech enhancement on VoiceBank test set 2 · VoiceBank test set 2 | CSIG: 3.99 -> 4.38 | Improved |
| use the RDL-Net rather than Metric-GAN | speech enhancement on VoiceBank test set 2 · VoiceBank test set 2 | CBAK: 3.18 -> 3.43 | Improved |
| use the RDL-Net rather than Metric-GAN | speech enhancement on VoiceBank test set 2 · VoiceBank test set 2 | COVL: 3.42 -> 3.72 | Improved |
| use the RDL-Net rather than Metric-GAN | speech enhancement on VoiceBank test set 2 · VoiceBank test set 2 | PESQ: 2.86 -> 3.02 | Improved |
| use the RDL-Net rather than MMSE-GAN | speech enhancement on VoiceBank test set 2 · VoiceBank test set 2 | STOI: 93 -> 94 | Improved |
| add local residual links (ablation) | RDL-Net without LR or GD · RDL-Net validation | validation error: 146.37 -> 142.14 | Improved |
| add global dense links (ablation) | RDL-Net without LR or GD · RDL-Net validation | validation error: 146.37 -> 141.82 | Improved |
| add both local residual links and global dense links (ablation) | RDL-Net without LR or GD · RDL-Net validation | validation error: 146.37 -> 141.43 | Improved |
| add global dense links without local residual links (ablation) | RDL-Net · RDL-Net validation | validation error: 141.82; validation-error fluctuations reported | Mixed |
| use RDL blocks instead of residual blocks | speech enhancement network at approximately 2 million parameters · Test set 1, street music, 10 dB | MOS-LQO: +0.22 MOS-LQO | Improved |
| use RDL blocks instead of residual blocks | speech enhancement network at approximately 2 million parameters · Test set 1, factory noise, 0 dB | STOI: 3.5% | Improved |
Closest papers by meaning
- Residual Dense Network for Image Restoration (TPAMI 2020 · similarity 0.87)
- Interactive Speech and Noise Modeling for Speech Enhancement (AAAI 2021 · similarity 0.84)
- Phase-Aware Speech Enhancement with Deep Complex U-Net (ICLR 2019 · similarity 0.83)
- Neural Residual Diffusion Models for Deep Scalable Vision Generation (NeurIPS 2024 · similarity 0.83)
- FINALLY: fast and universal speech enhancement with studio-like quality (NeurIPS 2024 · similarity 0.82)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.81)
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.
