FreeSonic: Training-Free Temporal-Aware Decoupled Attention for Precise Audio Editing

2026-06-13

AI Summary

Existing text-to-audio generation methods struggle to precisely edit target segments while preserving temporal consistency and background content. This work proposes a training-free audio editing framework built upon the TangoFlux model, which leverages an inversion-reconstruction pipeline and a joint text-audio attention mechanism to accurately localize target regions. The approach innovatively introduces scheduled attention decoupling and task-oriented noise injection strategies to disentangle localized editing from contextual preservation. Without requiring fine-tuning, the method achieves high-fidelity, temporally coherent, and precise audio editing, significantly outperforming current state-of-the-art techniques across multiple editing tasks.

Abstract

Text-to-audio (TTA) generation has made significant strides, yet achieving precise and consistent audio editing remains a major challenge. However, existing methods struggle to balance temporal consistency with background preservation. In this paper, we propose FreeSonic, a training-free framework leveraging the state-of-the-art Rectified Flow-based TangoFlux model. FreeSonic utilizes an optimized inversion-reverse process and joint text-audio attention maps for precise target segment extraction. For content editing, a novel scheduled attention decoupling confines modifications to target regions while preserving original acoustic context. Furthermore, task-oriented noise injection enhances versatility for tasks such as audio removal and non-rigid replacement. Extensive experimental results demonstrate that FreeSonic achieves a superior balance by providing a high-fidelity and efficient solution for precise and consistent audio editing. Project and demos: https://free-sonic.github.io/

Problem

• audio editing
• temporal consistency
• background preservation
• text-to-audio generation

Innovation

• training-free
• temporal-aware attention
• attention decoupling
• audio editing
• noise injection