Saved in:
Bibliographic Details
Main Authors: Li, Zhaoqing, Xu, Haoning, Su, Jingran, Liu, Yaofang, Rao, Zhefan, Wang, Huimeng, Deng, Jiajun, Wang, Tianzi, Jin, Zengrui, Liu, Rui, Che, Haoxuan, Liu, Xunying
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.31530
Tags: Add Tag
No Tags, Be the first to tag this record!
Table of Contents:
  • We present UNISON, a latent diffusion framework that unifies speech generation, sound generation, and audio editing within a single model. A single model handles text-to-audio, text-to-speech, zero-shot speaker cloning, mixed speech-and-sound generation, scene-level audio editing, speech-in-scene editing, and timed temporal composition, all of which share a single set of weights. Our architecture features two core designs: (1) Layer-wise deep LLM fusion, which injects hidden states from uniformly sampled layers of a frozen MLLM into corresponding MM-DiT blocks via learned projections, providing depth-matched semantic conditioning that improves instruction following over single-layer baselines; and (2) a unified multi-task architecture where task identity is encoded solely by a channel-wise mask and source audio is provided through VAE-encoded channel concatenation. Training is stabilized by an online GPU-side multi-task data synthesis pipeline with task-homogeneous batching and a two-stage curriculum. With 621M--732M trainable parameters, UNISON achieves results competitive with or exceeding task-specialist models across evaluated domains, while being roughly $4\times$ smaller than comparable unified systems.