batchalign 0.7.3b6__tar.gz → 0.7.3b8__tar.gz

{batchalign-0.7.3b6/batchalign.egg-info → batchalign-0.7.3b8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.3b6
+Version: 0.7.3b8
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.7.3b6 → batchalign-0.7.3b8}/batchalign/document.py

@@ -198,6 +198,11 @@ class Utterance(BaseModel):
             t = self._detokenize()
 
         t = t.replace(". . .", "+...")
+        t = t.replace("¿", "").replace("¡", "")
+        t = re.sub(r"^\+\.\.\.", "", t.strip()).strip()
+        t = re.sub(r"^\W+", "", t.strip()).strip()
+        t = re.sub(r",", " , ", t.strip()).strip()
+        t = re.sub(r" +", " ", t.strip()).strip()
         return t
 
     def __repr__(self):

{batchalign-0.7.3b6 → batchalign-0.7.3b8}/batchalign/models/whisper/infer_asr.py

@@ -3,6 +3,7 @@ from torchaudio import load
 import numpy as np 
 import os
 
+import re
 from transformers import pipeline
 
 from dataclasses import dataclass
@@ -189,16 +190,38 @@ class WhisperASRModel(object):
             element = groups.pop(0)
 
             if element["type"] == "text":
-                text = {
+                pl = element["payload"].strip()
+                before = re.findall(r"^\W+", pl)
+                after = re.findall(r"\W+$", pl)
+                texts = []
+                if len(before) > 0:
+                    texts.append({
+                        "type": "punct",
+                        "ts": element["start"],
+                        "end_ts": element["end"] if element["end"] else element["start"]+1,
+                        "value": before[0],
+                    })
+                    pl = pl.strip(before[0])
+                if len(after) > 0:
+                    pl = pl.strip(after[0])
+                texts.append({
                     "type": "text",
                     "ts": element["start"],
                     "end_ts": element["end"] if element["end"] else element["start"]+1,
-                    "value": element["payload"].strip(),
-                }
+                    "value": pl.strip(),
+                })
+                if len(after) > 0:
+                    texts.append({
+                        "type": "punct",
+                        "ts": element["start"],
+                        "end_ts": element["end"] if element["end"] else element["start"]+1,
+                        "value": after[0],
+                    })
 
-                if text["ts"] != text["end_ts"] and text["value"].strip() != "…":
-                    # text with no DTW time is likely a spurious retrace
-                    current_turn.append(text)
+                for text in texts:
+                    if text["ts"] != text["end_ts"] and text["value"].strip() != "…" and text["value"].strip() != "":
+                        # text with no DTW time is likely a spurious retrace
+                        current_turn.append(text)
             elif element["type"] == "segment" and current_speaker != element["payload"]:
                 turns.append({
                     "elements": current_turn,

{batchalign-0.7.3b6 → batchalign-0.7.3b8}/batchalign/pipelines/asr/utils.py

@@ -25,6 +25,7 @@ def retokenize(intermediate_output):
         tmp = []
         for word, bullet in utterance:
             word = word.replace("。", ".")
+            word = word.replace("¿", " ").replace("¡", " ")
             tmp.append((word, bullet))
             if word in ENDING_PUNCT or word[-1] in ENDING_PUNCT:
                 if word in ENDING_PUNCT:
@@ -178,14 +179,20 @@ def process_generation(output, lang="eng", utterance_engine=None):
                            id=f"PAR{speaker}",
                            name=f"Participant")
         words = []
-        for word, (start,end) in utterance:
-            if word not in ENDING_PUNCT:
+        for indx, (word, (start,end)) in enumerate(utterance):
+            if indx == 0:
+                seen_word = False
+            if word.strip() == "":
+                continue
+            if word not in ENDING_PUNCT+MOR_PUNCT:
                 if start == None or end == None:
                     words.append(Form(text=word, time=None))
                 else:
+                    seen_word = True
                     words.append(Form(text=word, time=(int(start), int(end))))
             else:
-                words.append(Form(text=word, time=None))
+                if seen_word:
+                    words.append(Form(text=word, time=None))
 
         final_utterances.append(Utterance(
             tier=participant,

batchalign-0.7.3b8/batchalign/version

@@ -0,0 +1,3 @@
+0.7.3-beta.8
+June 14, 2024
+more asr fixes

{batchalign-0.7.3b6 → batchalign-0.7.3b8/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.3b6
+Version: 0.7.3b8
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

batchalign-0.7.3b6/batchalign/version

@@ -1,3 +0,0 @@
-0.7.3-beta.6
-June 11, 2024
-hebrew unicode patches