Replace cited large-v3 baseline with measured 65.93 WER / 37.68 CER

README.md

@@ -49,18 +49,16 @@ Read that one if you want to know how these numbers were produced.
 
 ## Results
 
-Official Vaani Bhojpuri test split, n = 1,426. All rows measured on the same
-utterances with the same normalizer, except where noted.
+Official Vaani Bhojpuri test split, n = 1,426. Every row measured on the same
+utterances, with the same normalizer and the same metric implementation.
 
 | Model | Params | WER ↓ | CER ↓ |
 |---|---|---|---|
 | `openai/whisper-small` (untrained) | 244M | 122.07 | 78.75 |
-| `openai/whisper-large-v3` (untrained) † | 1.55B | 66.06 | 37.71 |
+| `openai/whisper-large-v3` (untrained) | 1.55B | 65.93 | 37.68 |
 | `ARTPARK-IISc/SraVaani` | 430M | **34.80** | 20.29 |
 | **This model** | 244M | 36.41 | **17.05** |
 
-† Cited from the project spec, not measured here.
-
 Against SraVaani — the model trained by the dataset's own authors — this is a
 split decision, and both halves are statistically significant under a paired
 bootstrap: SraVaani is better on WER by 1.61 (95% CI [+0.83, +2.44]), this