DRAMA-Eval - Prosody (Chinese): leaderboard
Speech generators voice whole multi-speaker scenes from structured Chinese scripts built from real podcasts, films and TV series (a set of 537 cases, each turn with delivery directions; systems were scored on up to that many); a Gemini 3.8 Flash judge compares each turn's prosody and fluency with the real recording, and the score is a tie-adjusted win rate where 50 means as good as the recording. Unlike EmergentTTS-Eval, the reference is the original human performance, not another TTS system.
Metric: Tie-adjusted win rate (%) against the real recording on prosody and fluency of each turn, Gemini 3.8 Flash A/B judge (50 = parity), Chinese set (537 cases). Source: aslp-lab.github.io. Saturation forecast: Rough model projection: around 2026. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | StepAudio 3 Gen | 58.3 | |
| 2 | MiniMax-H3 | 57.9 | |
| 3 | SeedAudio 1.0 | 57.4 | |
| 4 | Ming-3.0-tts-0.5B | 55.6 | |
| 5 | FireRedTTS3 | 55.1 | |
| 6 | Qwen3-TTS-12Hz-1.7B-VoiceDesign | 54.6 | |
| 7 | Ming-3.0-tts-tiny | 49.5 | |
| 8 | VoxCPM2 | 48.6 | |
| 9 | MOSS-TTS-2.0 Preview | 44.5 | |
| 10 | FireRedTTS2 | 39.5 |
Interactive version: theaggregate.ai/benchmark?slug=drama-eval-prosody-chinese · How It Works · Data refreshed daily, snapshot 2026-11-13.