Research map / Image generation / GANs
Chunked Autoregressive GAN 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 chunked autoregression instead of non-autoregressive waveform generation | HiFi-GAN-style conditional waveform synthesis · VCTK | Pitch error (cents): 51.2 -> 29.4 | Helped |
| use chunked autoregression instead of non-autoregressive waveform generation | HiFi-GAN-style conditional waveform synthesis · DAPS | Periodicity RMSE: .142 -> .107 | Helped |
| use chunked autoregression instead of non-autoregressive waveform generation | HiFi-GAN-style conditional waveform synthesis · VCTK | Training speed (ms/step): 1186 -> 502 | Helped |
| use chunked autoregression instead of non-autoregressive waveform generation | HiFi-GAN-style conditional waveform synthesis · VCTK synthetic cumulative sum, 1024 | L1 distance: .050 -> .009 | Helped |
| replace the GAN-TTS generator with the HiFi-GAN generator (ablation) | CARGAN · VCTK | Pitch error (cents): 29.4 -> 37.9 | Hurt |
| replace the GAN-TTS generator with the HiFi-GAN generator (ablation) | CARGAN · DAPS | Pitch error (cents): 21.6 -> 27.0 | Hurt |
| remove loss compensation and restore the original loss weights (ablation) | CARGAN · VCTK | Pitch error (cents): 29.4 -> 33.7 | Hurt |
| remove loss compensation and restore the original loss weights (ablation) | CARGAN · DAPS | Pitch error (cents): 21.6 -> 34.1 | Hurt |
| omit prepending autoregressive context to discriminator inputs (ablation) | CARGAN · VCTK | Periodicity RMSE: .086 -> .088 | Hurt |
| omit prepending autoregressive context to discriminator inputs (ablation) | CARGAN · VCTK | Pitch error (cents): 29.4 -> 24.6 | Helped |
| vary chunk size (ablation) | CARGAN · DAPS | Pitch error (cents): 22.9 (512); 22.1 (1024); 21.6 (2048); 24.8 (4096); 24.1 (8192) | Mixed |
| increase every convolution kernel size to 15 (ablation) | CARGAN · VCTK speech synthesis | Pitch error (cents): 29.4 -> 24.3 | Helped |
Closest papers by meaning
- MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis (NeurIPS 2019 · similarity 0.93)
- GANSynth: Adversarial Neural Audio Synthesis (ICLR 2019 · similarity 0.92)
- Symbolic Music Generation with Transformer-GANs (AAAI 2021 · similarity 0.90)
- PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation (ICLR 2025 · similarity 0.87)
- Fast Timing-Conditioned Latent Audio Diffusion (ICML 2024 · similarity 0.87)
- 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.