Seed-TTS-eval
Qwen3-TTS-12Hz-1.7B-Base 0.77% Models clone a voice from a short prompt and speak new text in English and Mandarin. Scored by word error rate from an ASR model and by speaker similarity.
Uses 1,000 Common Voice samples for English and 2,000 DiDiSpeech-2 samples for Mandarin, plus a hard Mandarin set. Repo has about 1,600 GitHub stars.
- Made by
- ByteDance (Seed-TTS team)
- Released
- June 2024
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Open-ended
- Tasks
- All public
- Last in a launch post
- March 2026, dropped before it saturated
- Last model added
- April 2026
Qwen3-TTS-12Hz-1.7B-Base leads with 0.77%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
- 1 Qwen3-TTS-12Hz-1.7B-Base 0.77%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Qwen3-TTS-12Hz-1.7B-Base | 0.77% | Jan 2026 | Lab model card |
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs headline it less often now, last in March 2026
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Mar 30, 2026 | Qwen3.5-Omni-Plus | 1.35 | Seed-en custom voice stability (Seed-zh 1.07, Seed-hard 6.24) |
| Jan 22, 2026 | Qwen3-TTS-12Hz-1.7B-Base | 0.77 | SEEDtest-zh WER, Qwen3-TTS-12Hz-1.7B-Base |
| Sep 22, 2025 | Qwen3-Omni-30B-A3B | 1.39 | SEED test-zh / test-en 1.07 / 1.39 (test-en recorded) |
| Mar 27, 2025 | Qwen2.5-Omni-7B | 2.32 | SEED test-en content consistency, Qwen2.5-Omni-7B_RL (test-zh 1.42, test-hard 6.54) |
5 system cards report it, from 3 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Apr 17, 2026 | Qwen3.5-Omni-Plus technical report | SEED zero-shot TTS WER test-zh 0.99, test-en 1.26, no overall (Table 8) | Capabilities | |
| Feb 4, 2026 |  ERNIE 5.0 technical report | SEED-TTS test-zh 1.35, test-en 1.54, WER, lower is better, no overall score (Table 8) | Capabilities | |
| Sep 22, 2025 | Qwen3-Omni-30B-A3B technical report | SEED test-zh 1.07, test-en 1.39, content consistency WER, lower is better, no overall score (Table 13) | Capabilities | |
| Sep 18, 2025 | MiMo-Audio-7B-Instruct technical report | Seed-TTS-Eval WER ZH 1.96, EN 5.37, ZH-Hard 14.14, lower is better, no overall score (Table 9) | Capabilities | |
| Mar 26, 2025 | Qwen2.5-Omni-7B technical report | SEED content consistency WER, Qwen2.5-Omni-7B RL test-zh 1.42, test-en 2.33, test-hard 6.54 (ICL 1.70, 2.72, 7.97). Speaker similarity RL 0.754, 0.641, 0.752. Lower WER is better, no overall score (Table 9) | Capabilities |
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy. Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7% - MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy. Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy. Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00% - VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules. Qwen3.5-Omni-Plus 93.1%