PyPI - batchalign - Versions diffs - 0.7.5a4__tar.gz → 0.7.5a5__tar.gz - Mend

@@ -19,8 +19,8 @@ class EvaluationEngine(BatchalignEngine):
     @staticmethod
     def __compute_wer(doc, gold):
         # get the text of the document and get the text of the gold
-        forms = [ j.text.lower() for i in doc.content for j in i.content ]
-        gold_forms = [ j.text.lower() for i in gold.content for j in i.content ]
+        forms = [ j.text.lower() for i in doc.content for j in i.content if isinstance(i, Utterance)]
+        gold_forms = [ j.text.lower() for i in gold.content for j in i.content if isinstance(i, Utterance)]
         # dp!
         alignment = align(forms, gold_forms, False)
@@ -61,8 +61,15 @@ class EvaluationEngine(BatchalignEngine):
             else:
                 prev_error = None
+        diff = []
+        for i in alignment:
+            if isinstance(i, Extra):
+                diff.append(f"{'+' if i.extra_type == ExtraType.REFERENCE else '-'} {i.key}")
+            else:
+                diff.append(f"  {i.key}")
         # wer = (S+D+I)/N
-        return (sub+dl+ins)/len(gold_forms)
+        return (sub+dl+ins)/len(gold_forms), "\n".join(diff)
     def analyze(self, doc, **kwargs):
         gold = kwargs.get("gold")
@@ -71,8 +78,11 @@ class EvaluationEngine(BatchalignEngine):
         if not gold or not isinstance(gold, Document):
             raise ValueError(f"Unexpected format for gold transcript. Expected batchalign.Document, got '{type(gold)}'")
+        wer, diff = self.__compute_wer(doc, gold)
         return {
-            "wer": self.__compute_wer(doc, gold)
+            "wer": wer,
+            "diff": diff
         }

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a4
+Version: 0.7.5a5
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

@@ -272,6 +272,9 @@ def benchmark(ctx, in_dir, out_dir, lang, num_speakers, whisper, **kwargs):
         # write the wer
         with open(Path(output).with_suffix(".wer.txt"), 'w') as df:
             df.write(str(doc["wer"]))
+        with open(Path(output).with_suffix(".diff"), 'w') as df:
+            df.write(str(doc["diff"]))
     _dispatch("benchmark", lang, num_speakers, ["mp3", "mp4", "wav"], ctx,
               in_dir, out_dir,

@@ -0,0 +1,3 @@
+0.7.5-alpha.5
+September 3nd, 2024
+fix benchmark command, part 2

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a4
+Version: 0.7.5a5
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

@@ -1,3 +0,0 @@
-0.7.5-alpha.4
-September 3nd, 2024
-fix benchmark command

batchalign 0.7.5a4__tar.gz → 0.7.5a5__tar.gz

batchalign 0.7.5a4tar.gz → 0.7.5a5tar.gz