Update FAA-Word-Model/README.md

FAA-Word-Model/README.md

@@ -7,7 +7,7 @@ Proposed CRNN with Frequency-Adaptive Attention between BiLSTM layers 2 and 3.
 - CNN backbone: 6 convolutional blocks (max 256 channels)
 - Recurrent block: 3 BiLSTM layers, hidden = 160 per direction
 - Decoding: Connectionist Temporal Classification (CTC)
-- Parameters: 4,000,923
+- Parameters: 3,997,859
 
 ## Test Performance (DASTNUS, seed 42)
 | Metric | Value |