Research map / Image generation / Fast diffusion sampling
WaveEx: Accelerating Flow Matching-based Speech Generation via Wavelet-guided Extrapolation
Xiaoqian Liu, Xiyan Gui, Zhengkun Ge, Yuan Ge, Chang Zou, Jiacheng Liu, Zhikang Niu, Qixi Zheng, Chen Xu, Xie Chen, Tong Xiao, Jingbo Zhu and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 two-phase schedule instead of uniform extrapolation schedule (ablation) | F5-TTS + WaveEx · LibriSpeech-PC | WER (%): 5.697 -> 2.175 | Improved |
| use two-phase schedule instead of quadratic schedule (ablation) | F5-TTS + WaveEx · LibriSpeech-PC | WER (%): 3.138 -> 2.175 | Improved |
| extrapolate high-frequency band instead of zeroing it (ablation) | DiffRhythm + WaveEx · DiffRhythm music | SongEval scores | Improved |
| extrapolate high-frequency band instead of freezing it (ablation) | DiffRhythm + WaveEx · DiffRhythm music | SongEval scores | Improved |
| first-order instead of second-order Taylor extrapolation (ablation) | F5-TTS + WaveEx · F5-TTS latent trajectory | trajectory prediction accuracy | Improved |
| add WaveEx to 6-NFE F5-TTS | F5-TTS + WaveEx · LibriSpeech-PC | WER (%): 4.323 -> 2.175 | Improved |
| add WaveEx to 6-NFE E2-TTS | E2-TTS + WaveEx · LibriSpeech-PC | WER (%): 20.141 -> 3.323 | Improved |
| WaveEx instead of EPSS on F5-TTS | F5-TTS + WaveEx · LibriSpeech-PC | WER (%): 2.227 vs 2.175 | No clear effect |
| WaveEx instead of FORA/TaylorSeer/ToCa cache methods | F5-TTS + WaveEx · LibriSpeech-PC | WER (%): 2.751-3.015 -> 2.175 | Improved |
| add WaveEx to 6-NFE FlowSE | FlowSE + WaveEx · DNS-Challenge 2023 | OVRL: 2.774 -> 2.868 | Improved |
Closest papers by meaning
- PeriodWave: Multi-Period Flow Matching for High-Fidelity Waveform Generation (ICLR 2025 · similarity 0.88)
- Generative Pre-training for Speech with Flow Matching (ICLR 2024 · similarity 0.86)
- UniFlow-Audio: Unified Flow Matching for Audio Generation from Omni-Modalities (NeurIPS 2026 · similarity 0.86)
- Shallow Flow Matching for Coarse-to-Fine Text-to-Speech Synthesis (NeurIPS 2025 · similarity 0.85)
- DiffAR: Denoising Diffusion Autoregressive Model for Raw Speech Waveform Generation (ICLR 2024 · similarity 0.85)
- Wavelet Flow Matching for Multi-Scale Physics Emulation (NeurIPS 2026 · 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.
