mict-zhaw commited on
Commit
f98f1b9
1 Parent(s): 29ac49f

Update `num_to_words`, `filter_and_clean_text` & `preprocess_text` and add `map_special_terms`

Browse files
Files changed (1) hide show
  1. chall.py +211 -13
chall.py CHANGED
@@ -1,12 +1,13 @@
1
  import json
2
  import os
3
- from typing import Union, List
 
 
4
  from datasets import DatasetInfo, BuilderConfig, GeneratorBasedBuilder, Version, Features, Value, Audio, SplitGenerator, Split, logging
5
  from datasets.features import Sequence
6
  import soundfile as sf
7
  import importlib.util
8
 
9
-
10
  _SAMPLE_RATE = 16000
11
 
12
  _DESCRIPTION = "tbd"
@@ -22,17 +23,32 @@ class ChallConfig(BuilderConfig):
22
  split_segments: bool = False
23
 
24
  # settings that can only be used together with split_segments
25
- max_chunk_length: Union[float, None] = None
26
- min_chunk_length: Union[float, None] = None
27
- max_pause_length: Union[float, None] = None
28
  remove_trailing_pauses: bool = False
29
 
 
 
 
 
 
30
  def __init__(self, **kwargs):
31
- self.split_segments = kwargs.pop("split_segments", ChallConfig.split_segments)
32
- self.remove_trailing_pauses = kwargs.pop("remove_trailing_pauses", ChallConfig.remove_trailing_pauses)
33
- self.max_chunk_length = kwargs.pop("max_chunk_length", ChallConfig.max_chunk_length)
34
- self.min_chunk_length = kwargs.pop("min_chunk_length", ChallConfig.min_chunk_length)
35
- self.max_pause_length = kwargs.pop("max_pause_length", ChallConfig.max_pause_length)
 
 
 
 
 
 
 
 
 
 
36
 
37
  super(ChallConfig, self).__init__(**kwargs)
38
 
@@ -66,6 +82,8 @@ class Chall(GeneratorBasedBuilder):
66
  max_chunk_length=12,
67
  min_chunk_length=0.5,
68
  remove_trailing_pauses=True,
 
 
69
  description="Settings used for the paper."
70
  )
71
  ]
@@ -86,7 +104,6 @@ class Chall(GeneratorBasedBuilder):
86
  """
87
  self._check_dependencies()
88
  super().__init__(**kwargs)
89
- print(self.config)
90
 
91
  @staticmethod
92
  def _check_dependencies() -> None:
@@ -122,6 +139,8 @@ class Chall(GeneratorBasedBuilder):
122
  "area_of_school_code": Value("int32"),
123
  "background_noise": Value("bool"),
124
  "speaker": Value("string"),
 
 
125
  "words": Sequence(
126
  {
127
  "start": Value("float"),
@@ -138,6 +157,8 @@ class Chall(GeneratorBasedBuilder):
138
  "intervention": Value("int32"),
139
  "school_grade": Value("string"),
140
  "area_of_school_code": Value("int32"),
 
 
141
  "participants": Sequence(
142
  {
143
  "pseudonym": Value("string"),
@@ -243,8 +264,7 @@ class Chall(GeneratorBasedBuilder):
243
  else:
244
  yield from self._generate_transcript_examples(audio_id, str(audio_file_path), row, transcript)
245
 
246
- @staticmethod
247
- def _generate_transcript_examples(audio_id: str, audio_file_path: str, data: dict, transcript: dict):
248
  """
249
  Generates examples based on the entire audio file and its associated transcript metadata. This method reads the
250
  entire audio file, extracts speaker and segment information from the transcript, and packages these along with
@@ -261,6 +281,9 @@ class Chall(GeneratorBasedBuilder):
261
  transcript_data["speakers"] = transcript.get("speakers", [])
262
  transcript_data["segments"] = transcript.get("segments", [])
263
 
 
 
 
264
  with sf.SoundFile(audio_file_path) as audio_file:
265
  audio = audio_file.read(dtype='float32')
266
 
@@ -295,6 +318,9 @@ class Chall(GeneratorBasedBuilder):
295
  segment_data["speaker_id"] = segment["speaker"]
296
  segment_data["words"] = segment["words"]
297
 
 
 
 
298
  start_time = segment["words"][0]["start"]
299
  end_time = segment["words"][-1]["end"]
300
  start_frame = int(_SAMPLE_RATE * start_time)
@@ -443,4 +469,176 @@ class Chall(GeneratorBasedBuilder):
443
 
444
  return list_of_chunks
445
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
446
 
 
 
1
  import json
2
  import os
3
+ import re
4
+ import string
5
+ from typing import Union, List, Dict
6
  from datasets import DatasetInfo, BuilderConfig, GeneratorBasedBuilder, Version, Features, Value, Audio, SplitGenerator, Split, logging
7
  from datasets.features import Sequence
8
  import soundfile as sf
9
  import importlib.util
10
 
 
11
  _SAMPLE_RATE = 16000
12
 
13
  _DESCRIPTION = "tbd"
 
23
  split_segments: bool = False
24
 
25
  # settings that can only be used together with split_segments
26
+ max_chunk_length: Union[float, None]
27
+ min_chunk_length: Union[float, None]
28
+ max_pause_length: Union[float, None]
29
  remove_trailing_pauses: bool = False
30
 
31
+ lowercase: bool
32
+ num_to_words: bool
33
+ allowed_chars: set
34
+ special_terms_mapping: dict
35
+
36
  def __init__(self, **kwargs):
37
+ self.split_segments = kwargs.pop("split_segments", False)
38
+ self.remove_trailing_pauses = kwargs.pop("remove_trailing_pauses", False)
39
+
40
+ self.max_chunk_length = kwargs.pop("max_chunk_length", None)
41
+ self.min_chunk_length = kwargs.pop("min_chunk_length", None)
42
+ self.max_pause_length = kwargs.pop("max_pause_length", None)
43
+
44
+ self.lowercase = kwargs.pop("lowercase", True)
45
+ self.num_to_words = kwargs.pop("num_to_words", True)
46
+ self.special_terms_mapping = kwargs.pop("special_terms_mapping", {})
47
+
48
+ if self.lowercase:
49
+ self.allowed_chars = set(string.ascii_lowercase + " äöü'")
50
+ else:
51
+ self.allowed_chars: set = set(string.ascii_lowercase + string.ascii_uppercase + " ÄÖÜäöü'")
52
 
53
  super(ChallConfig, self).__init__(**kwargs)
54
 
 
82
  max_chunk_length=12,
83
  min_chunk_length=0.5,
84
  remove_trailing_pauses=True,
85
+ lowercase=True,
86
+ num_to_words=True,
87
  description="Settings used for the paper."
88
  )
89
  ]
 
104
  """
105
  self._check_dependencies()
106
  super().__init__(**kwargs)
 
107
 
108
  @staticmethod
109
  def _check_dependencies() -> None:
 
139
  "area_of_school_code": Value("int32"),
140
  "background_noise": Value("bool"),
141
  "speaker": Value("string"),
142
+ "raw_text": Value("string"),
143
+ "clear_text": Value("string"),
144
  "words": Sequence(
145
  {
146
  "start": Value("float"),
 
157
  "intervention": Value("int32"),
158
  "school_grade": Value("string"),
159
  "area_of_school_code": Value("int32"),
160
+ "raw_text": Value("string"),
161
+ "clear_text": Value("string"),
162
  "participants": Sequence(
163
  {
164
  "pseudonym": Value("string"),
 
264
  else:
265
  yield from self._generate_transcript_examples(audio_id, str(audio_file_path), row, transcript)
266
 
267
+ def _generate_transcript_examples(self, audio_id: str, audio_file_path: str, data: dict, transcript: dict):
 
268
  """
269
  Generates examples based on the entire audio file and its associated transcript metadata. This method reads the
270
  entire audio file, extracts speaker and segment information from the transcript, and packages these along with
 
281
  transcript_data["speakers"] = transcript.get("speakers", [])
282
  transcript_data["segments"] = transcript.get("segments", [])
283
 
284
+ transcript_data["raw_text"] = raw_text = self.get_raw_text([word for segment in transcript["segments"] for word in segment["words"]])
285
+ transcript_data["clear_text"] = self.get_clear_text(raw_text)
286
+
287
  with sf.SoundFile(audio_file_path) as audio_file:
288
  audio = audio_file.read(dtype='float32')
289
 
 
318
  segment_data["speaker_id"] = segment["speaker"]
319
  segment_data["words"] = segment["words"]
320
 
321
+ segment_data["raw_text"] = raw_text = self.get_raw_text(segment["words"])
322
+ segment_data["clear_text"] = self.get_clear_text(raw_text)
323
+
324
  start_time = segment["words"][0]["start"]
325
  end_time = segment["words"][-1]["end"]
326
  start_frame = int(_SAMPLE_RATE * start_time)
 
469
 
470
  return list_of_chunks
471
 
472
+ @staticmethod
473
+ def get_raw_text(words: List[Dict]) -> str:
474
+ """
475
+
476
+ """
477
+ raw_text = " ".join([word["text"] for word in words])
478
+ return raw_text
479
+
480
+ def get_clear_text(self, raw_text: str) -> str:
481
+ """
482
+ Processes the raw text to produce a clear, cleaned version by removing annotations,
483
+ preprocessing the text, converting numbers to words, mapping special terms,
484
+ converting to lowercase, and filtering allowed characters.
485
+
486
+ :param raw_text: The raw input text to be processed.
487
+ :return: A string representing the processed clear text.
488
+ """
489
+
490
+ clear_text = self.remove_annotations(raw_text)
491
+ clear_text = self.preprocess_text(clear_text)
492
+
493
+ if self.config.num_to_words:
494
+ clear_text = self.num_to_words(clear_text)
495
+
496
+ if self.config.special_terms_mapping:
497
+ self.map_special_terms(clear_text, special_terms_mapping=self.config.special_terms_mapping)
498
+
499
+ if self.config.lowercase:
500
+ clear_text = clear_text.lower()
501
+
502
+ if self.config.allowed_chars:
503
+ clear_text = self.filter_and_clean_text(clear_text, allowed_chars=self.config.allowed_chars)
504
+
505
+ return clear_text
506
+
507
+ @staticmethod
508
+ def preprocess_text(transcript: str) -> str:
509
+ """
510
+ Preprocesses the text by removing words between brackets and parentheses,
511
+ standardizing spaces before apostrophes, removing commas between digits,
512
+ and replacing special characters.
513
+
514
+ :param transcript: The input transcript to preprocess.
515
+ :return: The preprocessed transcript with various text normalization applied.
516
+ """
517
+
518
+ transcript = re.sub(r"[<\[][^>\]]*[>\]]", "", transcript) # remove words between brackets
519
+ transcript = re.sub(r"\(([^)]+?)\)", "", transcript) # remove words between parenthesis
520
+ transcript = re.sub(r"\s+'", "'", transcript) # standardize when there's a space before an apostrophe
521
+ transcript = re.sub(r"(\d),(\d)", r"\1\2", transcript) # remove commas between digits
522
+ transcript = re.sub(r"\.([^0-9]|$)", r" \1", transcript) # remove periods not followed by numbers
523
+
524
+ # Replace special characters
525
+ special_chars = {
526
+ 'ß': 'ss', 'ç': 'c', 'á': 'a', 'à': 'a', 'â': 'a', 'é': 'e', 'è': 'e', 'ê': 'e', 'í': 'i', 'ì': 'i', 'î': 'i',
527
+ 'ó': 'o', 'ò': 'o', 'ô': 'o', 'ú': 'u', 'ù': 'u', 'û': 'u', '-': ' ', '\u2013': ' ', '\xad': ' ', '/': ' '
528
+ }
529
+ for char, replacement in special_chars.items():
530
+ transcript = transcript.replace(char, replacement)
531
+
532
+ # Normalize whitespace
533
+ transcript = re.compile(r'[ \t]+').sub(' ', transcript)
534
+
535
+ return transcript
536
+
537
+ @staticmethod
538
+ def remove_annotations(transcript: str) -> str:
539
+ """
540
+ Removes specific annotations and conventions from the transcript
541
+
542
+ :param transcript: The transcript to preprocess.
543
+ :return: The preprocessed transcript with conventions and annotations removed.
544
+ """
545
+ transcript = transcript.replace('@g', '') # (Swiss-)German words
546
+ transcript = transcript.replace('@?', '') # best guess
547
+ transcript = transcript.replace('@!', '') # Errors
548
+ transcript = transcript.replace('-', '') # Repetitions
549
+ transcript = transcript.replace('--', '') # Reformulations
550
+ transcript = transcript.replace('(...)', '') # Long pauses
551
+ transcript = transcript.replace('(Whispering)', '') # Whispering
552
+ transcript = transcript.replace('(whispers)', '') # Whispering
553
+ transcript = transcript.replace('(whispering)', '') # Whispering
554
+ transcript = transcript.replace('(unv.)', '') # ?
555
+ transcript = transcript.replace('(laughing)', '') # Laughing
556
+ transcript = transcript.replace('(laughs)', '') # Laughing
557
+ transcript = transcript.replace('(Laughter)', '') # Laughing
558
+ return transcript
559
+
560
+ @staticmethod
561
+ def num_to_words(transcript: str) -> str:
562
+ """
563
+ Converts numerical expressions in the transcript to their word equivalents using the num2words library.
564
+
565
+ :param transcript: The input transcript containing numerical expressions.
566
+ :return: The transcript with numerical expressions converted to words.
567
+ """
568
+
569
+ from num2words import num2words
570
+
571
+ def replace(match):
572
+ number_str = match.group(0)
573
+ if re.match(r'\d+\.', number_str):
574
+ # Check if this is an ordinal context by looking at the following character
575
+ next_char_index = match.end()
576
+ if next_char_index < len(transcript) and transcript[next_char_index].islower():
577
+ # Convert to ordinal if followed by a lowercase letter
578
+ number = int(number_str[:-1]) # Remove the period
579
+ return num2words(number, to='ordinal')
580
+ else:
581
+ # Treat as the end of a sentence, return as is
582
+ return number_str
583
+
584
+ elif re.match(r'\d{4}s', number_str):
585
+ # Convert decades
586
+ number = int(number_str[:-1])
587
+ return num2words(number, to='year') + "s"
588
+
589
+ elif re.match(r'\d+m\b', number_str):
590
+ # Convert numbers with 'm' (meters) suffix
591
+ number = int(number_str[:-1])
592
+ return num2words(number) + " meters"
593
+
594
+ elif number_str[-2:] in ['st', 'nd', 'rd', 'th']:
595
+ # Convert ordinal numbers with suffix
596
+ number = int(re.match(r'\d+', number_str).group(0))
597
+
598
+ return num2words(number, to='ordinal')
599
+
600
+ else:
601
+ # Convert cardinal numbers
602
+ return num2words(number_str)
603
+
604
+ # Regular expression to find numbers, ordinals, ordinals with period, decades, and numbers with 'm' suffix
605
+ pattern = re.compile(r'\b\d+(\.\d+)?\b|\b\d+(st|nd|rd|th)\b|\b\d+\.\b|\b\d{4}s\b|\b\d+m\b')
606
+
607
+ # Substitute numbers with their word equivalent
608
+ new_sentence = pattern.sub(replace, transcript)
609
+
610
+ return new_sentence
611
+
612
+ @staticmethod
613
+ def map_special_terms(transcript: str, special_terms_mapping: dict):
614
+ """
615
+ Maps special terms in the transcript to their corresponding replacements using dictionary of pairs
616
+
617
+ :param transcript: The input transcript containing special terms to be mapped.
618
+ :param special_terms_mapping: A dictionary where keys are special terms and values are their replacements.
619
+ :return: The transcript with special terms replaced.
620
+ """
621
+
622
+ for term, replacement in special_terms_mapping.items():
623
+ transcript = re.sub(r'\b' + re.escape(term) + r'\b', replacement, transcript, flags=re.IGNORECASE)
624
+ return transcript
625
+
626
+ @staticmethod
627
+ def filter_and_clean_text(transcript: str, allowed_chars: set = None):
628
+ """
629
+ Filters the transcript to include only the allowed characters and normalizes
630
+ whitespace by removing extra spaces and trimming the text.
631
+
632
+ :param transcript: The input transcript to be filtered and cleaned.
633
+ :param allowed_chars: A set of allowed characters. If provided, only these characters will be retained in the transcript.
634
+ :return: The filtered and cleaned transcript.
635
+ """
636
+
637
+ # Filter allowed characters
638
+ if allowed_chars is not None:
639
+ transcript = ''.join([char for char in transcript if char in allowed_chars])
640
+
641
+ # Normalize whitespace
642
+ transcript = re.compile(r'[ \t]+').sub(' ', transcript).strip()
643
 
644
+ return transcript