Lune

ACL Findings2026

Bypassing Neural Evaluations for Fast Audio Editing via Adaptive Trajectory Extrapolation

Xiaoqian Liu, Zhengkun Ge, Jianjin Wang, Haoran Zhang, Yuan Ge, Kaiyan Chang, Chen Xu, Tong Xiao, Zhengtao Yu, Linfeng Zhang, JingBo Zhu

摘要

Recent advancements in audio diffusion models have significantly improved text-to-audio editing via inversion techniques. However, these models typically rely on dense, fixedstep sampling trajectories to maintain structural integrity during inversion and generation, leading to prohibitive computational costs. We propose AdaTE, a model-agnostic Adaptive Trajectory Extrapolation framework that accelerates the inversion-based editing process by dynamically evaluating only the most critical generative phases. Specifically, we introduce a hierarchical probing mechanism that monitors curvature acceleration and information gain to detect pivotal transitions within the latent flow. This allows the model to selectively skip redundant segments via linear extrapolation while preserving dense neural evaluations for complex semantic changes. Extensive experiments across AudioLDM2, Auffusion, and Tango2 demonstrate that AdaTE achieves up to a 3.9ˆspeedup with negligible loss in fidelity. AdaTE significantly shifts the Pareto frontier, providing an efficient solution for high-fidelity audio synthesis and editing.

问 Lune

几分钟读透这篇 Paper,不用再花一下午。

Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。

可以从这些问题问起

Lune 可以查阅

全文 · 相关的 Paper

获取答案

免费开始,无需绑卡

脉络

  • Consistency Models 引用了 Denoising Diffusion Implicit Models
  • Consistency Models 引用了 DiffWave: A Versatile Diffusion Model for Audio Synthesis
  • DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps 引用了 Denoising Diffusion Implicit Models
  • SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations 引用了 Denoising Diffusion Implicit Models
它引用的引用它的从左到右按年份排列,引用越多,位置越高。

它引用的 Paper19

  1. Denoising Diffusion Implicit ModelsJiaming Song, Chenlin Meng, Stefano ErmonICLR 2021 · 11,743 次引用
  2. DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 StepsCheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen 等NeurIPS 2022 · 2,653 次引用
  3. SDEdit: Guided Image Synthesis and Editing with Stochastic Differential EquationsChenlin Meng, Yutong He, Yang Song, Jiaming Song 等ICLR 2022 · 2,128 次引用
  4. DiffWave: A Versatile Diffusion Model for Audio SynthesisZhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao 等ICLR 2021 · 1,902 次引用
  5. Consistency ModelsYang Song, Prafulla Dhariwal, Mark Chen, Ilya SutskeverICML 2023 · 1,720 次引用

相关 Paper

  1. WaveEx: Accelerating Flow Matching-based Speech Generation via Wavelet-guided ExtrapolationXiaoqian Liu, Xiyan Gui, Zhengkun Ge, Yuan Ge 等AAAI 2026 · 1 次引用
  2. Diffusion Adaptive Text Embedding for Text-to-Image Diffusion ModelsByeonghu Na, Minsang Park, Gyuwon Sim, Donghyeok Shin 等NeurIPS 2025 · 8 次引用
  3. IMPACT: Iterative Mask-based Parallel Decoding for Text-to-Audio Generation with Diffusion ModelingKuan-Po Huang, Shu-Wen Yang, Huy Phan, Bo-Ru Lu 等ICML 2025
  4. Melodia: Training-Free Music Editing Guided by Attention Probing in Diffusion ModelsYi Yang, Haowen Li, Tianxiang Li, Boyu Cao 等AAAI 2026 · 1 次引用
  5. Text-to-Audio Generation using Instruction Guided Latent Diffusion ModelDeepanway Ghosal, Navonil Majumder, Ambuj Mehrish, Soujanya PoriaACM MM 2023 · 90 次引用
黄昏的海面,两侧是细线勾勒的悬崖