STEB - Translation COMET (Chinese to English): leaderboard
Chinese to English speech-to-speech translation on STEB, a 32.6-hour benchmark of real drama, audiobook, advertisement, interview, news and commentary speech (Cheng et al., 2026). Translation fidelity: COMET (wmt22-comet-da, 0-1) of the ASR transcript of the translated speech against the reference translation; higher is better.
Metric: COMET (wmt22-comet-da, 0-1 scale) of the Qwen3-ASR transcript of the translated speech against the reference translation; Chinese-to-English speech-to-speech translation. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Step-Audio-2 | 0.78 | |
| 2 | UniSS | 0.77 | |
| 3 | Seed LiveInterpret 2.0 | 0.76 | |
| 4 | SeamlessExpressive | 0.69 |
Interactive version: theaggregate.ai/benchmark?slug=steb-translation-comet-chinese-to-english · How It Works · Data refreshed daily, snapshot 2026-11-12.