Research map / Large language models / Language, safety and interpretability
CoCoEmo: Composable and Controllable Human-Like Emotional TTS via Activation Steering
Siyi Wang, Shihong Tan, Siyi Liu, Hong Jia, Gongping Huang, James Bailey, Ting Dang
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 |
|---|---|---|---|
| drive emotion via SLM instead of flow-matching module (ablation) | CosyVoice2 · cross-conditioning | F0 CCC: 0.305 -> 0.109 | Improved |
| drive emotion via SLM instead of flow-matching module (ablation) | CosyVoice2 · cross-conditioning | Energy CCC: 0.737 -> 0.308 | Improved |
| drive emotion via SLM instead of flow-matching module (ablation) | CosyVoice2 · cross-conditioning | SR Std: 0.518 -> 0.691 | Improved |
| add steering (alpha=5.0) to CosyVoice2 | CosyVoice2 · CREMA-D mixed emotion | TEP: 0.065 -> 0.149 | Improved |
| add steering on top of Instruction1 | CosyVoice2 · CREMA-D mixed emotion | TEP: 0.200 -> 0.335 | Improved |
| add steering on top of Emo-Vector | IndexTTS2 · CREMA-D mixed emotion | Spearman rho: 0.236 -> 0.399 | Improved |
| add steering (alpha=5.0) to CosyVoice2 | CosyVoice2 · IEMOCAP mixed emotion (OOD) | TEP: 0.197 -> 0.253 | Improved |
| add steering on top of Emo-Vector (alpha=5.0) | IndexTTS2 · IEMOCAP mixed emotion (OOD) | E-SIM: 0.855 -> 0.824 | Hurt |
| add steering on top of Instruction1 (alpha=3.0) | CosyVoice2 · IEMOCAP mixed emotion (OOD) | N-MOS: 3.88 -> 2.18 | Hurt |
| raise alpha from 3.0 to 6.0 (ablation) | CosyVoice2 · IEMOCAP high mismatch | TEP: 0.332 -> 0.504 | Improved |
| add steering (alpha=6.0) over Instruction | CosyVoice2 · IEMOCAP high mismatch | TEP: 0.436 -> 0.504 | Improved |
| raise alpha from 3.0 to 5.0 (ablation) | CosyVoice2 · CREMA-D mixed emotion | TEP: 0.100 -> 0.149 | Improved |
Closest papers by meaning
- Word-Level Emotional Expression Control in Zero-Shot Text-to-Speech Synthesis (NeurIPS 2025 · similarity 0.89)
- EmergentTTS-Eval: Evaluating TTS Models on Complex Prosodic, Expressiveness, and Linguistic Challenges Using Model-as-a-Judge (NeurIPS 2025 · similarity 0.86)
- Emotional Face-to-Speech (ICML 2025 · similarity 0.85)
- Emotion Rendering for Conversational Speech Synthesis with Heterogeneous Graph-Based Context Modeling (AAAI 2024 · similarity 0.85)
- EmotionThinker: Prosody-Aware Reinforcement Learning for Explainable Speech Emotion Reasoning (ICLR 2026 · similarity 0.84)
- Emotional Speech-driven 3D Body Animation via Disentangled Latent Diffusion (CVPR 2024 · 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.
