Research map / Multi-agent systems and games / Human-AI and goal-conditioned
AgentSteerTTS: A Multi-Agent Closed-Loop Framework for Composite-Instruction Text-to-Speech
Bin Kang, Shaoguo Wen, Yang Fan, Shunlong Wu, Junjie Wang, Yulin Li, Junzhi Zhao, Junle Wang, Zhuotao Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add adversarial disentanglement (ablation) | AgentSteerTTS · Composite-instruction benchmark | Delta S-SIM: 0.048 -> 0.021 | Improved |
| add adversarial disentanglement (ablation) | AgentSteerTTS · Composite-instruction benchmark | S-SIM: 0.807 -> 0.841 | Improved |
| add adversarial disentanglement (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: 0.61 -> 0.78 | Improved |
| add prototype retrieval (ablation) | AgentSteerTTS · Composite-instruction benchmark | E-SIM: 0.910 -> 0.955 | Improved |
| add prototype retrieval (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: 0.49 -> 0.78 | Improved |
| add perceptual cropping of retrieved prototype (ablation) | AgentSteerTTS · Composite-instruction benchmark | Leakage: 0.21 -> 0.14 | Improved |
| add fast loop (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: 0.70 -> 0.78 | Improved |
| add slow loop (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: 0.67 -> 0.78 | Improved |
| add fast loop (ablation) | AgentSteerTTS · Composite-instruction benchmark | E-SIM: 0.951 -> 0.955 | No clear effect |
| increase prototype library from 25h to 100h (ablation) | AgentSteerTTS · Composite-instruction benchmark | E-SIM: 0.912 -> 0.955 | Improved |
| increase Fast Loop steps T from 0 to 2 (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: 0.70 -> 0.78 | Improved |
| increase Fast Loop steps T from 3 to 5 (ablation) | AgentSteerTTS · Composite-instruction benchmark | CSR: +0.01 | No clear effect |
Closest papers by meaning
- GROOT-2: Weakly Supervised Multimodal Instruction Following Agents (ICLR 2025 · similarity 0.85)
- Enhancing Speech Large Language Models through Reinforced Behavior Alignment (NeurIPS 2026 · similarity 0.85)
- Improving Instruction-Following in Language Models through Activation Steering (ICLR 2025 · similarity 0.85)
- EmergentTTS-Eval: Evaluating TTS Models on Complex Prosodic, Expressiveness, and Linguistic Challenges Using Model-as-a-Judge (NeurIPS 2025 · similarity 0.85)
- Word-Level Emotional Expression Control in Zero-Shot Text-to-Speech Synthesis (NeurIPS 2025 · similarity 0.84)
- $\tau$-Voice: Benchmarking Full-Duplex Voice Agents on Real-World Domains (ICML 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
