Research map / Image generation / Autoregressive and latent generation
STAR-VAE: Structured Topology-Aware Regularization for Audio Reconstruction and Generation
Huadai Liu, Wen Wang, Kaicheng Luo, Qian Chen, Xiangang Li, Wei Xue
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 |
|---|---|---|---|
| add STAR regularization (ablation) | hybrid CNN-Mamba VAE · AudioCaps reconstruction | FAD: 2.74 -> 2.31 | Improved |
| add STAR regularization (ablation) | CNN VAE · AudioCaps reconstruction | FAD: 3.36 -> 2.65 | Improved |
| add Mamba to a VAE without STAR (ablation) | CNN-VAE (isotropic KL) · AudioCaps reconstruction | STFT-D: 1.28 -> 1.35 | Hurt |
| add Mamba on top of STAR (ablation) | CNN-STAR · Song Describer reconstruction | FAD: 0.38 -> 0.25 | Improved |
| replace Mamba with Transformer (ablation) | STAR-VAE · Song Describer reconstruction | FAD: 0.25 -> 0.30 | Hurt |
| use STAR-VAE instead of SAO VAE | Stable Audio Open · AudioCaps text-to-audio | FDopenl3: 89.2 -> 72.5 | Improved |
| use STAR-VAE instead of SAO VAE (ablation) | STAR-Gen · AudioCaps text-to-audio | FDopenl3: 67.4 -> 55.8 | Improved |
| use STAR-VAE instead of ϵar-VAE (ablation) | STAR-Gen · AudioCaps text-to-audio | FDopenl3: 76.45 -> 55.8 | Improved |
| use STAR instead of isotropic KL at same 21.5Hz rate | Stable Audio Open VAE · AudioCaps reconstruction | FAD: 3.29 -> 2.31 | Improved |
| use power-law Gamma-Growth (gamma 2.0) instead of step or linear penalty (ablation) | STAR · AudioCaps/Song Describer reconstruction | reconstruction | Improved |
Closest papers by meaning
- Generative Audio Language Modeling with Continuous-valued Tokens and Masked Next-Token Prediction (ICML 2025 · similarity 0.86)
- V2A-Mapper: A Lightweight Solution for Vision-to-Audio Generation by Connecting Foundation Models (AAAI 2024 · similarity 0.85)
- Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis (ICLR 2025 · similarity 0.85)
- WavTokenizer: an Efficient Acoustic Discrete Codec Tokenizer for Audio Language Modeling (ICLR 2025 · similarity 0.85)
- SSC-VAE: Structured Sparse Coding Based Variational Autoencoder for Detail Preserved Image Reconstruction (AAAI 2025 · similarity 0.85)
- CV-VAE: A Compatible Video VAE for Latent Generative Video Models (NeurIPS 2024 · similarity 0.85)
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.
