DRAMA-Eval - Script Fidelity CER (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); the score is the character error rate of an ASR transcript of the generated audio against the script. Unlike Seed-TTS-Eval's read-aloud sentences, the text is multi-turn dialogue voiced as one scene.
Metric: Character error rate (%) of Qwen3-0.6B-ASR transcripts of each generated scene against its script, 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 | Ming-3.0-tts-0.5B | 2.55 | |
| 2 | SeedAudio 1.0 | 2.58 | |
| 3 | FireRedTTS3 | 2.64 | |
| 4 | Qwen3-TTS-12Hz-1.7B-VoiceDesign | 2.73 | |
| 5 | FireRedTTS2 | 3.01 | |
| 6 | VoxCPM2 | 3.25 | |
| 7 | StepAudio 3 Gen | 3.78 | |
| 8 | MiniMax-H3 | 3.84 | |
| 9 | MOSS-TTS-2.0 Preview | 4.28 | |
| 10 | Ming-3.0-tts-tiny | 4.67 |
Interactive version: theaggregate.ai/benchmark?slug=drama-eval-script-fidelity-cer-chinese · How It Works · Data refreshed daily, snapshot 2026-11-13.