AI
EmoRES-TTS: Residual-Enhanced Vector Steering for Emotional Speech Generation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 07:23:06 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 07:18:24 UTC.
Briefings
This study introduces EmoRES, a training-free vector steering method for emotional speech generation that decomposes emotion vectors into shared and residual components. On IEMOCAP, EmoRES significantly outperforms CoCoEmo across four objective metrics, achieving a 26.13 percentage point rank correlation improvement and a 35.0% relative gain in listener emotion identification. Human evaluation further showed up to a 17.3% improvement in fidelity and a 63.8% higher preference for naturalness compared to CoCoEmo. The method preserves the shared component while strengthening the residual component to achieve effective control without retraining the backbone.
Key points
- EmoRES decomposes emotion vectors into a shared component that moves speech away from neutral expression and a residual component that directs generation toward the requested emotion.
- On IEMOCAP, EmoRES outperforms CoCoEmo across all four objective emotion metrics, with rank correlation improving by 26.13 percentage points on IndexTTS-2 and 12.97 percentage points on CosyVoice2.
Context and limitations
The study covers only two backbones (IndexTTS-2 and CosyVoice2) and four objective emotion metrics; human evaluation results are based on pairwise comparisons and may vary across different listeners or tasks.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 07:23:06 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidjg2ktkdwbjdz5uerxqxpwgussgeqplje4hv7ogwdlmyucddb4qi