Research map / Image generation / Autoregressive and latent generation
A Spectral Energy Distance for Parallel Speech 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 |
|---|---|---|---|
| train with spectral GED loss instead of adversarial loss (same generator) | GAN-TTS generator · TTS validation | cFDSD: 0.077 -> 0.038 | Helped |
| train with spectral GED loss instead of adversarial loss (same generator) | GAN-TTS generator · TTS test sentences | MOS: 4.16 -> 4.03 | Hurt |
| add repulsive term to GED (ablation) | GED · TTS test sentences | MOS: 3.00 -> 4.03 | Helped |
| add repulsive term to GED (ablation) | GED · TTS validation (200k steps) | cFDSD: 0.047 -> 0.040 | Helped |
| use multi-scale spectrogram windows instead of a single window (ablation) | GED · TTS validation (200k steps) | cFDSD: 0.043 -> 0.040 | Helped |
| increase DCT/DST overcompleteness beyond 8x (ablation) | GED · TTS validation (200k steps) | cFDSD: 0.040 -> 0.041 | No clear effect |
| add unconditional GAN loss to GED | GED + GAN-TTS generator · TTS test sentences | MOS: 4.03 -> 4.25 | Helped |
| add GED loss to GAN-TTS (full conditional discriminators) (ablation) | GAN-TTS · TTS test sentences | MOS: 4.16 -> 4.24 | Helped |
| replace conditional discriminators with GED loss in GED+GAN (ablation) | GED + GAN · TTS test sentences | MOS: 4.24 -> 4.25 | No clear effect |
| replace GAN-TTS generator with iSTFT generator under GED | GED · TTS test sentences | MOS: 4.03 -> 4.10 | No clear effect |
Closest papers by meaning
- High Fidelity Speech Synthesis with Adversarial Networks (ICLR 2020 · similarity 0.91)
- Efficient Speech Language Modeling via Energy Distance in Continuous Latent Space (NeurIPS 2025 · similarity 0.87)
- PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation (ICLR 2025 · similarity 0.86)
- Metis: A Foundation Speech Generation Model with Masked Generative Pre-training (NeurIPS 2025 · similarity 0.86)
- Continuous Autoregressive Modeling with Stochastic Monotonic Alignment for Speech Synthesis (ICLR 2025 · similarity 0.85)
- DiffAR: Denoising Diffusion Autoregressive Model for Raw Speech Waveform Generation (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.