Research map / Image generation / GANs
Avocodo: Generative Adversarial Network for Artifact-Free Vocoder
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 |
|---|---|---|---|
| Replace HiFi-GAN V1 with Avocodo V1 | GAN-based vocoder · LJSpeech test | F0 RMSE: 35.96 -> 33.98 | Helped |
| Replace HiFi-GAN V1 with Avocodo V1 | GAN-based vocoder · LJSpeech test | LSD-HF: 9.72 -> 9.13 | Helped |
| Replace HiFi-GAN V1 with Avocodo V1 | GAN-based vocoder · LJSpeech test | PESQ: 3.64 -> 3.81 | Helped |
| Replace HiFi-GAN V1 with Avocodo V1 | GAN-based vocoder · Unseen(EN) | MOS: 3.940(±0.09) -> 4.080(±0.08) | Helped |
| Replace HiFi-GAN V1 with Avocodo V1 | GAN-based vocoder · Unseen(KR) | MOS: 3.810(±0.06) -> 3.972(±0.07) | Helped |
| Use CoMBD instead of MSD (ablation) | Avocodo generator V2 · LJSpeech single-speaker comparison | MOS: 3.743 (±0.07) -> 4.156 (±0.05) | Helped |
| Use SBD instead of MPD (ablation) | Avocodo generator V2 · LJSpeech single-speaker comparison | MOS: 3.675 (±0.08) -> 4.130 (±0.06) | Helped |
| Use PQMF downsampling instead of naive downsampling | GAN-based vocoder training · High-pitch singing voice | F0 reconstruction | Helped |
| Add collaborative multi-scale and hierarchical discrimination of intermediate and downsampled outputs | Avocodo generator · Speech waveform synthesis | upsampling artifacts | Helped |
| Discriminate PQMF-decomposed sub-band signals with SBD (ablation) | Avocodo generator · LJSpeech single-speaker comparison | MOS: 3.675 (±0.08) -> 4.130 (±0.06) | Helped |
Closest papers by meaning
- BigVGAN: A Universal Neural Vocoder with Large-Scale Training (ICLR 2023 · similarity 0.90)
- Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (ICLR 2024 · similarity 0.88)
- FINALLY: fast and universal speech enhancement with studio-like quality (NeurIPS 2024 · similarity 0.85)
- FlowDec: A flow-based full-band general audio codec with high perceptual quality (ICLR 2025 · similarity 0.83)
- DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024 · similarity 0.83)
- PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation (ICLR 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.