STEB - Emotion Preservation (Chinese to English): leaderboard
Metric: LLM-judged emotion consistency (1-5) of category, intensity and attitude between the source and the translated speech: both are captioned by Qwen3-Omni-Captioner and summarized by Qwen3-30B-A3B, which scores the pair three times (outlier-aware aggregate); Chinese-to-English speech-to-speech translation of the normal subset's Chinese source utterances. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | UniSS | 3.61 |
| 2 | Step-Audio-2 | 3.57 |
| 3 | Seed LiveInterpret 2.0 | 3.34 |
| 4 | SeamlessExpressive | 3.1 |
Interactive version: theaggregate.ai/benchmark?slug=steb-emotion-preservation-chinese-to-english · How It Works · Data refreshed daily, snapshot 2026-09-26.