Drop datasets: field — ZeroBench-TTS is eval data, not training data

README.md

@@ -15,8 +15,12 @@ tags:
   - voice-cloning
   - vietnamese-tts
   - tieng-viet
-datasets:
-  - zeroweight-ai/ZeroBench-TTS
+# NOTE deliberately NO `datasets:` field. It is the only thing that populates
+# the Hub's cross-link, but the Hub renders it as "Models trained or fine-tuned
+# on <dataset>" — which for our own held-out benchmark reads as train/test
+# contamination. ZeroBench-TTS is EVALUATION data; every voice in it is held
+# out of training. The `model-index` block below states that correctly, and the
+# body links the benchmark in prose.
 metrics:
   - wer
 model-index:
@@ -133,12 +137,13 @@ for why both matter.
 
 | | **ZeroTTS** | XTTS-v2-vietnamse | viXTTS |
 |---|:-:|:-:|:-:|
-| **WER** ↓ | **1.03%** | 16.42% | 18.40% |
+| **WER — raw text** ↓ | **1.03%** | 16.42% | 18.40% |
+| **WER — pre-normalized text** ↓ | **0.56%** | 7.27% | 8.61% |
 | **Naturalness** (UTMOS) ↑ | **2.91** | 2.43 | 2.35 |
 | **Voice similarity** (SSIM) ↑ | 0.936 | **0.940** | 0.935 |
 | **Dead air** (excess silence) ↓ | **0.029 s** | 0.532 s | 0.233 s |
 
-**16× fewer word errors**, ~0.5 MOS more natural, an order of magnitude less
+**16× fewer word errors** on the real task, **13× fewer** even after handing every model a perfect text frontend; ~0.5 MOS more natural, an order of magnitude less
 dead air. Median WER is **0.00%** on all four subsets.
 
 WER by subset: