通过剪枝采样步骤，让语音合成模型提速4倍且音质不变。
Accelerating Flow-Matching-Based Text-to-Speech via Empirically Pruned Step Sampling
- 基于实际采样轨迹，剔除冗余步骤，采用非均匀采样策略。
- 7步生成，推理实时因子0.030，比原模型快4倍。
- 无需训练，可通用至其他流匹配语音模型，适合高效部署场景。
基于流匹配的文本到语音模型（如Voicebox、E2 TTS和F5-TTS）近年来受到广泛关注。这些模型需多步采样从噪声重建语音，导致推理速度成为关键瓶颈。减少采样步数可显著提升效率。为此，本文提出Fast F5-TTS，一种无需训练的加速推理方法。通过分析F5-TTS的采样轨迹，识别出冗余步骤，提出经验剪枝步采样（EPSS），一种非均匀时间步采样策略，有效减少采样步数。该方法在NVIDIA RTX 3090 GPU上实现7步生成，推理实时因子（RTF）为0.030，比原始F5-TTS快4倍，同时保持相当的生成质量。此外，EPSS在E2 TTS模型上也表现良好，展现出强泛化能力。
原文摘要 · Abstract (English)
Flow-matching-based text-to-speech (TTS) models, such as Voicebox, E2 TTS, and F5-TTS, have attracted significant attention in recent years. These models require multiple sampling steps to reconstruct speech from noise, making inference speed a key challenge. Reducing the number of sampling steps can greatly improve inference efficiency. To this end, we introduce Fast F5-TTS, a training-free approach to accelerate the inference of flow-matching-based TTS models. By inspecting the sampling trajectory of F5-TTS, we identify redundant steps and propose Empirically Pruned Step Sampling (EPSS), a non-uniform time-step sampling strategy that effectively reduces the number of sampling steps. Our approach achieves a 7-step generation with an inference RTF of 0.030 on an NVIDIA RTX 3090 GPU, making it 4 times faster than the original F5-TTS while maintaining comparable performance. Furthermore, EPSS performs well on E2 TTS models, demonstrating its strong generalization ability.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。