PyPI - batchalign - Versions diffs - 0.7.10.post4__tar.gz → 0.7.11a0__tar.gz - Mend

@@ -107,11 +107,20 @@ class WhisperFAEngine(BatchalignEngine):
             # we do this BACKWARDS because we went to have the first timestamp
             # we get about a word first
             alignments.reverse()
-            for elem in alignments:
+            for indx,elem in enumerate(alignments):
                 if isinstance(elem, Match):
+                    next_elem = indx - 1 # remember this is backwards, see above
+                    while next_elem >= 0 and alignments[next_elem].payload == elem.payload:
+                        next_elem -= 1
+                    if next_elem < 0:
+                        next_elem = None
+                    else:
+                        next_elem = alignments[next_elem]
                     grp[elem.reference_payload][0].time = (int(round((timings[elem.payload]*1000 +
                                                                       grp[0][1][0]))),
                                                            int(round((timings[elem.payload]*1000 +
+                                                                      grp[0][1][0])))+500 if next_elem == None else
+                                                           int(round((timings[next_elem.payload]*1000 +
                                                                       grp[0][1][0]))))
         L.debug(f"Correcting text...")
@@ -144,8 +153,9 @@ class WhisperFAEngine(BatchalignEngine):
                             w.time = (w.time[0], doc.content[next_ut].alignment[0])
                         else:
                             w.time = (w.time[0], w.time[0]+500) # give half a second because we don't know
-                    else:
-                        w.time = (w.time[0], ut.content[tmp].time[0])
+                    # else:
+                    #     w.time = (w.time[0], ut.content[tmp].time[0])
                     # just in case, bound the time by the utterance derived timings
                     if ut.alignment and ut.alignment[0] != None:
                         w.time = (max(w.time[0], ut.alignment[0]), min(w.time[1], ut.alignment[1]))

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.10.post4
+Version: 0.7.11a0
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

@@ -82,7 +82,7 @@ class WhisperFAModel(object):
     def __call__(self, audio, text):
         L.debug("Whisper Preprocessing...")
         # input features
-        features = self.__processor(audio=audio, text=text,
+        features = self.__processor(audio=audio, text=" ".join(list(text)),
                                     sampling_rate=self.sample_rate,
                                     return_tensors='pt')
         tokens = features["labels"][0]

@@ -990,7 +990,8 @@ def morphoanalyze(doc: Document, retokenize:bool, status_hook:callable = None, *
                     content.dependency = form.dependency
         except Exception as e:
-            warnings.warn(f"Utterance failed parsing, skipping ud tagging... line='{line}', error='{e}'.\n")
+            pass
+            # warnings.warn(f"Utterance failed parsing, skipping ud tagging... line='{line}', error='{e}'.\n")
     L.debug("Stanza done.")
     return doc

@@ -0,0 +1,3 @@
+0.7.11-alpha.0
+Janurary 22th, 2025
+Whisper inference with spaces

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.10.post4
+Version: 0.7.11a0
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

@@ -1,3 +0,0 @@
-0.7.10-post.4
-Janurary 28th, 2025
-numbers

batchalign 0.7.10.post4__tar.gz → 0.7.11a0__tar.gz

batchalign 0.7.10.post4tar.gz → 0.7.11a0tar.gz