Research map / Image generation / GANs
MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis
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 |
|---|---|---|---|
| use weight normalization instead of spectral normalization (ablation) | MelGAN generator · LJ Speech mel-spectrogram inversion | MOS: 1.33 -> 3.09 | Helped |
| avoid adding L1 loss in audio space (ablation) | MelGAN · LJ Speech mel-spectrogram inversion | MOS: 2.59 -> 3.09 | Helped |
| use a window-based discriminator (ablation) | MelGAN · LJ Speech mel-spectrogram inversion | MOS: 2.29 -> 3.09 | Helped |
| add dilated convolutions to the generator (ablation) | MelGAN · LJ Speech mel-spectrogram inversion | MOS: 2.60 -> 3.09 | Helped |
| use a multi-scale rather than single discriminator (ablation) | MelGAN · LJ Speech mel-spectrogram inversion | MOS: 2.93 -> 3.09 | Helped |
| use weight normalization (ablation) | MelGAN · LJ Speech mel-spectrogram inversion | MOS: 3.03 -> 3.09 | Helped |
| replace WaveGlow with MelGAN as the Text2mel vocoder | Text2mel TTS · End-to-end TTS | MOS: 4.10 -> 3.72 | Hurt |
| replace Griffin-Lim with MelGAN for mel inversion | MelGAN · VCTK mel-spectrogram inversion | MOS: 1.72 -> 3.49 | Helped |
Closest papers by meaning
- Chunked Autoregressive GAN for Conditional Waveform Synthesis (ICLR 2022 · similarity 0.93)
- GANSynth: Adversarial Neural Audio Synthesis (ICLR 2019 · similarity 0.91)
- Adversarial Audio Synthesis (ICLR 2019 · similarity 0.91)
- HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis (NeurIPS 2020 · similarity 0.90)
- DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024 · similarity 0.86)
- Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (ICLR 2024 · similarity 0.86)
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.