Support subword and word language models

Closes #199 (closed)

Ref https://redmine.teklia.com/issues/4941

dan/ocr/decoder.py

@@ -571,10 +571,14 @@ class CTCLanguageDecoder:
             ).strip()
             for hypothesis in hypotheses
         ]
-        # Normalize confidence score
+        # Normalize confidence scoref"{np.around(np.mean(char_confidences[current : next_token]), 2)}
         out["confidence"] = [
-            np.exp(
-                hypothesis[0].score / ((self.language_model_weight + 1) * length.item())
+            np.around(
+                np.exp(
+                    hypothesis[0].score
+                    / ((self.language_model_weight + 1) * length.item())
+                ),
+                2,
             )
             for hypothesis, length in zip(hypotheses, batch_sizes)
         ]