FdAudio: MeanFlow-Anchored Fréchet-Distance Post-Training for One-Step Text-to-Audio Generation

2026-07-11

AI Summary

Current one-step text-to-audio generation models significantly underperform multi-step approaches in audio quality. This work proposes a post-training method in the embedding space that directly optimizes the distribution of one-step generation by introducing a multi-representation Fréchet Distance (FD) loss, while incorporating a MeanFlow consistency constraint as a structural anchor to effectively prevent performance degradation relative to multi-step models. The approach achieves, for the first time, compatibility between high-quality one-step and multi-step audio generation: compared to the MeanAudio baseline, it reduces the FD score by 11.4% and improves the Fréchet Audio Distance (FAD) by 28.8%. Notably, it matches or exceeds the fidelity of state-of-the-art multi-step models using only 25 sampling steps, substantially reducing computational latency.

Abstract

While recent few-step sampling text-to-audio generation models like MeanAudio substantially accelerate generation by modeling average velocities, their strict one-step generation quality still lags significantly behind multi-step counterparts. We propose FdAudio to bridge this gap. Unlike MeanAudio, which relies solely on regression against target velocity fields, our post-training approach optimizes the final one-step distribution directly across pre-trained embedding spaces via a multi-representation Fréchet-distance (FD) loss. Crucially, to prevent the multi-step degradation that naive post-training with FD-loss causes, we introduce a MeanFlow consistency objective as a structural anchor. Results demonstrate that FdAudio establishes state-of-the-art one-step T2A generation quality among few-step systems, yielding an 11.4% reduction in FD score and a 28.8% improvement in FAD score relative to the baseline MeanAudio framework. Notably, we solve FD post-training's naive multi-step degradation issue by proposing the MeanFlow anchor, enabling a 25-step sampling path to maintain high-fidelity audio synthesis that matches or surpasses strong multi-step models at a fraction of their computational latency.

Problem

• text-to-audio generation
• one-step sampling
• generation quality
• Fréchet distance
• multi-step degradation

Innovation

• Fréchet Distance
• One-Step Generation
• Post-Training
• MeanFlow Consistency
• Text-to-Audio