Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.17583 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916021111095296 |
|---|---|
| author | Kang, Bin Wen, Shaoguo Fan, Yang Wu, Shunlong Wang, Junjie Li, Yulin Zhao, Junzhi Wang, Junle Tian, Zhuotao |
| author_facet | Kang, Bin Wen, Shaoguo Fan, Yang Wu, Shunlong Wang, Junjie Li, Yulin Zhao, Junzhi Wang, Junle Tian, Zhuotao |
| contents | While existing text-to-speech (TTS) models exhibit high expressiveness, fine-grained control over composite instructions remains challenging due to the structural mismatch between discrete textual intents and continuous acoustic realizations. Inspired by human cognitive decoupling, we introduce AgentSteerTTS, a multi-agent closed-loop framework designed for intent-faithful expressive control of composite instructions. First, in our framework, an adversarial disentanglement agent mitigates speaker-emotion leakage by learning separable identity and emotion-prosody subspaces with leakage-suppressing regularization. Next, a Dual-Stream Anchoring Controller grounds abstract intents using a large-scale acoustic prototype library: a Retrieval Agent selects expressive anchors, while a Synthesis Agent fuses them into continuous control vectors via gated attention. Finally, a Fast-Slow Feedback Agent refines output intensity through latent gradient correction and resolves semantic-acoustic mismatches using high-level perceptual critique. Experiments on a composite-instruction benchmark and public test sets show that AgentSteerTTS yields consistent and significant improvements to the baselines, demonstrating the effectiveness of the proposed method. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_17583 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | AgentSteerTTS: A Multi-Agent Closed-Loop Framework for Composite-Instruction Text-to-Speech Kang, Bin Wen, Shaoguo Fan, Yang Wu, Shunlong Wang, Junjie Li, Yulin Zhao, Junzhi Wang, Junle Tian, Zhuotao Computer Vision and Pattern Recognition While existing text-to-speech (TTS) models exhibit high expressiveness, fine-grained control over composite instructions remains challenging due to the structural mismatch between discrete textual intents and continuous acoustic realizations. Inspired by human cognitive decoupling, we introduce AgentSteerTTS, a multi-agent closed-loop framework designed for intent-faithful expressive control of composite instructions. First, in our framework, an adversarial disentanglement agent mitigates speaker-emotion leakage by learning separable identity and emotion-prosody subspaces with leakage-suppressing regularization. Next, a Dual-Stream Anchoring Controller grounds abstract intents using a large-scale acoustic prototype library: a Retrieval Agent selects expressive anchors, while a Synthesis Agent fuses them into continuous control vectors via gated attention. Finally, a Fast-Slow Feedback Agent refines output intensity through latent gradient correction and resolves semantic-acoustic mismatches using high-level perceptual critique. Experiments on a composite-instruction benchmark and public test sets show that AgentSteerTTS yields consistent and significant improvements to the baselines, demonstrating the effectiveness of the proposed method. |
| title | AgentSteerTTS: A Multi-Agent Closed-Loop Framework for Composite-Instruction Text-to-Speech |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.17583 |