eval: add VisualEars golden test set results (3 datasets)

README.md

@@ -61,9 +61,10 @@ Measured by Rade with greedy CTC decoding (fp16) on two standard Persian test se
 | Test set | Clips | WER | CER |
 |---|---|---|---|
 | **FLEURS** `fa_ir` — read speech | 871 | 19.6 % | **4.4 %** |
+| **VisualEars golden** `fa` — curated (clean/farfield/obstructed) | 6,669 | 22.9 % | **4.3 %** |
 | **Common Voice 17.0** `fa` — crowd-sourced | 10,355 | 21.8 % | 17.8 % |
 
-> **CER is the more faithful metric for Persian.** Persian **WER** is inflated by orthographic/spacing variation (نیم‌فاصله/ZWNJ, affix spacing, compound spelling) that doesn't reflect actual mis-recognition — note FLEURS sits at **19.6 % WER but only 4.4 % CER**, i.e. most "word errors" are one-character spelling differences. On clean read speech (FLEURS) the model reaches **CER ≈ 4 %**; on noisier crowd-sourced audio (Common Voice — spontaneous speech, varied mics/accents, loan words) **CER ≈ 18 %**.
+> **CER is the more faithful metric for Persian.** Persian **WER** is inflated by orthographic/spacing variation (نیم‌فاصله/ZWNJ, affix spacing, compound spelling) that doesn't reflect actual mis-recognition — note FLEURS sits at **19.6 % WER but only 4.4 % CER**, i.e. most "word errors" are one-character spelling differences. On clean, well-curated speech (FLEURS, VisualEars) the model reaches **CER ≈ 4 %**, and it stays robust across recording conditions (VisualEars far-field 4.5 % / obstructed 4.2 % CER). On noisier crowd-sourced audio (Common Voice — spontaneous speech, varied mics/accents, loan words) CER rises to **≈ 18 %**.
 
 ## Speed & hardware