batchalign 0.7.5a5__tar.gz → 0.7.5a7__tar.gz

{batchalign-0.7.5a5/batchalign.egg-info → batchalign-0.7.5a7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a5
+Version: 0.7.5a7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.7.5a5 → batchalign-0.7.5a7}/batchalign/models/utterance/infer.py

@@ -35,6 +35,7 @@ class BertUtteranceModel(object):
         self.model.eval()
 
     def __call__(self, passage):
+        print(passage)
         # input passage words removed of all preexisting punctuation
         passage = passage.lower()
         passage = passage.replace('.','')
@@ -67,7 +68,8 @@ class BertUtteranceModel(object):
         prev_word_idx = None
 
         # for each word, perform the action
-        for indx, elem in enumerate(tokd.word_ids(0)):
+        wids = tokd.word_ids(0)
+        for indx, elem in enumerate(wids):
             # if its none, append nothing or if we have
             # seen it before, do nothing
             if elem is None or elem == prev_word_idx:
@@ -81,23 +83,31 @@ class BertUtteranceModel(object):
             # set the working variable
             w = input_tokenized[elem]
 
-            # perform the edit actions
-            if action == 1:
-                w = w[0].upper() + w[1:]
-            elif action == 2:
-                w = w+'.'
-            elif action == 3:
-                w = w+'?'
-            elif action == 4:
-                w = w+'!'
-            elif action == 5:
-                w = w+','
+            # fix one word hanging issue
+            will_action = False
+            if indx < len(wids)-2 and classified_targets[0][indx+1] > 0:
+                will_action = True
+
+            if not will_action:
+                # perform the edit actions
+                if action == 1:
+                    w = w[0].upper() + w[1:]
+                elif action == 2:
+                    w = w+'.'
+                elif action == 3:
+                    w = w+'?'
+                elif action == 4:
+                    w = w+'!'
+                elif action == 5:
+                    w = w+','
+
 
             # append
             res_toks.append(w)
 
         # compose final passage
         final_passage = self.tokenizer.convert_tokens_to_string(res_toks)
+        print(final_passage)
         try: 
             split_passage = sent_tokenize(final_passage)
         except LookupError:

{batchalign-0.7.5a5 → batchalign-0.7.5a7}/batchalign/pipelines/analysis/eval.py

@@ -8,7 +8,7 @@ from batchalign.pipelines.base import *
 from batchalign.pipelines.asr.utils import *
 from batchalign.utils.config import config_read
 
-from batchalign.utils.dp import align, ExtraType, Extra
+from batchalign.utils.dp import align, ExtraType, Extra, Match
 
 import logging
 L = logging.getLogger("batchalign")
@@ -22,6 +22,9 @@ class EvaluationEngine(BatchalignEngine):
         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)]
 
+        forms = [i for i in forms if i.strip() not in MOR_PUNCT+ENDING_PUNCT]
+        gold_forms = [i for i in gold_forms if i.strip() not in MOR_PUNCT+ENDING_PUNCT]
+
         # dp!
         alignment = align(forms, gold_forms, False)
 
@@ -35,8 +38,16 @@ class EvaluationEngine(BatchalignEngine):
         # ie: if we have <extra.payload> <extra.reference> +> substitution
         #     but if we have <extra.reference> <extra.reference> this is 2 insertions
 
+        cleaned_alignment = []
+
         for i in alignment:
+
             if isinstance(i, Extra):
+                if len(cleaned_alignment) > 0 and i.extra_type == ExtraType.REFERENCE and "name" in i.key and i.key[:4] != "name":
+                    cleaned_alignment.pop(-1)
+                    cleaned_alignment.append(Match(i.key, None, None))
+                    continue
+
                 if prev_error != None and prev_error != i.extra_type:
                     # this is a substitution: we have different "extra"s in
                     # reference vs. playload
@@ -61,6 +72,8 @@ class EvaluationEngine(BatchalignEngine):
             else:
                 prev_error = None
 
+            cleaned_alignment.append(i)
+
         diff = []
         for i in alignment:
             if isinstance(i, Extra):

batchalign-0.7.5a7/batchalign/version

@@ -0,0 +1,3 @@
+0.7.5-alpha.7
+September 7th, 2024
+batch hanging utterance bug

{batchalign-0.7.5a5 → batchalign-0.7.5a7/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.5a5
+Version: 0.7.5a7
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

batchalign-0.7.5a5/batchalign/version

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