Research map / Image generation / Fast diffusion sampling
FloWaveNet : A Generative Flow for Raw Audio
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 Gaussian IAF with FloWaveNet | Parallel audio synthesis on LJSpeech · LJSpeech test set | 5-scale MOS: 3.75$\pm$ 0.159 -> 3.95$\pm$ 0.154 | Helped |
| Use non-causal instead of causal dilated convolutions (ablation) | FloWaveNet · FloWaveNet | 5-scale MOS: 3.36 $\pm$ 0.134 -> 3.95 $\pm$ 0.154 | Helped |
| Train Gaussian IAF with only KL divergence rather than KL + frame loss (ablation) | Gaussian IAF · Test data | KL divergence: 0.134 -> 0.040 | Helped |
| Train Gaussian IAF with only frame loss rather than KL + frame loss (ablation) | Gaussian IAF · Test data | KL divergence: 0.134 -> 1.378 | Hurt |
| Remove the spectrogram frame loss from KL distillation (ablation) | Gaussian IAF · Generated speech | Audio realism | Hurt |
| Remove the KL divergence loss and use only spectrogram frame loss (ablation) | Gaussian IAF · Generated speech | Noise level | Hurt |
| Use a lower prior temperature, with 0.8 selected for FloWaveNet | FloWaveNet · LJSpeech audio generation | Perceptual audio quality | Mixed |
Closest papers by meaning
- DFlow: A Generative Model Combining Denoising AutoEncoder and Normalizing Flow for High Fidelity Waveform Generation (ICML 2024 · similarity 0.89)
- FlowDec: A flow-based full-band general audio codec with high perceptual quality (ICLR 2025 · similarity 0.86)
- BinauralFlow: A Causal and Streamable Approach for High-Quality Binaural Speech Synthesis with Flow Matching Models (ICML 2025 · similarity 0.86)
- UniFlow-Audio: Unified Flow Matching for Audio Generation from Omni-Modalities (NeurIPS 2026 · similarity 0.86)
- Fast Timing-Conditioned Latent Audio Diffusion (ICML 2024 · similarity 0.85)
- Generative Pre-training for Speech with Flow Matching (ICLR 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.