Research map / Image generation / GANs
GANSynth: Adversarial Neural Audio 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 |
|---|---|---|---|
| Represent audio with log-magnitude and instantaneous-frequency spectra rather than directly generated waveforms | GANSynth · NSynth | Human Eval (wins): 485 vs 216 | Helped |
| Estimate instantaneous frequency rather than phase (ablation) | Spectral GAN · NSynth | Human Eval (wins): 283 vs 203 | Helped |
| Increase STFT frequency resolution (ablation) | Spectral GAN · NSynth | NDB: 37.0 -> 29.3 | Helped |
| Increase STFT frequency resolution (ablation) | IF-Mel GAN · NSynth | Human Eval (wins): 479 -> 485 | Helped |
| Apply mel frequency scaling (ablation) | IF GAN · NSynth | Human Eval (wins): 283 -> 479 | Helped |
| Apply mel frequency scaling (ablation) | IF + H GAN · NSynth | Human Eval (wins): 308 -> 485 | Helped |
| Use progressive training instead of non-progressive training (ablation) | Waveform GAN · NSynth waveform generation | FID: 447 -> 375 | Helped |
| Use progressive training instead of non-progressive training (ablation) | Waveform GAN · NSynth waveform generation | IS: 14.8 -> 2.5 | Hurt |
| Use 16-bit mixture-of-logistics output rather than 8-bit categorical output (ablation) | WaveNet · NSynth | FID: 320 -> 656 | Hurt |
| Generate with an IF-Mel GAN rather than WaveNet | Audio generation · NSynth, four-second sample | generation latency: 1077.53 seconds -> 20 milliseconds | Helped |
| Use progressive rather than non-progressive spectral GAN training (ablation) | Spectral GAN · NSynth | sample quality | No clear effect |
Closest papers by meaning
- Adversarial Audio Synthesis (ICLR 2019 · similarity 0.93)
- Chunked Autoregressive GAN for Conditional Waveform Synthesis (ICLR 2022 · similarity 0.92)
- HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis (NeurIPS 2020 · similarity 0.92)
- High Fidelity Speech Synthesis with Adversarial Networks (ICLR 2020 · similarity 0.92)
- Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (ICLR 2024 · similarity 0.87)
- Fast Timing-Conditioned Latent Audio Diffusion (ICML 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.