Accelerating Flow-Matching-Based Text-to-Speech via Empirically Pruned Step Sampling

TL;DR

This paper introduces Fast F5-TTS, a training-free method that employs Empirically Pruned Step Sampling to significantly accelerate flow-matching-based TTS models, reducing sampling steps and inference time while preserving quality.

Contribution

The paper proposes a novel non-uniform sampling strategy, EPSS, that reduces sampling steps in flow-matching TTS models without retraining, improving inference speed.

Findings

Achieves 4x faster inference with 7-step generation on F5-TTS.

Maintains comparable speech quality despite reduced sampling steps.

Demonstrates strong generalization on E2 TTS models.

Abstract

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.