georgechang8 commited on
Commit
7215e45
·
verified ·
1 Parent(s): 545c772

Update README.md

Browse files

Added cleansing steps

Files changed (1) hide show
  1. README.md +118 -24
README.md CHANGED
@@ -1,24 +1,5 @@
1
  ---
2
  dataset_info:
3
- - config_name: clean
4
- features:
5
- - name: audio
6
- dtype:
7
- audio:
8
- sampling_rate: 16000
9
- - name: text
10
- dtype: string
11
- - name: id
12
- dtype: string
13
- - name: session_id
14
- dtype: string
15
- splits:
16
- - name: train
17
- num_bytes: 3354307483.0
18
- num_examples: 46583
19
- download_size: 3346711427
20
- dataset_size: 3354307483.0
21
- - config_name: default
22
  features:
23
  - name: audio
24
  dtype:
@@ -37,10 +18,6 @@ dataset_info:
37
  download_size: 3346820592
38
  dataset_size: 3341105554.6299996
39
  configs:
40
- - config_name: clean
41
- data_files:
42
- - split: train
43
- path: clean/train-*
44
  - config_name: default
45
  data_files:
46
  - split: train
@@ -53,10 +30,33 @@ configs:
53
 
54
  This dataset is derived from espnet/yodas, more details can be found here: https://huggingface.co/datasets/espnet/yodas
55
 
56
- This is a subset of the zh000 subset of espnet/yodas dataset, which focuses on videos with Mandarin-English code-switching phenomenon.
 
 
57
 
58
  ## Dataset Details
59
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  ### Dataset Description
61
 
62
  <!-- Provide a longer summary of what this dataset is. -->
@@ -226,6 +226,100 @@ audio_dataset.push_to_hub(
226
  embed_external_files=True
227
  )
228
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
229
 
230
  ## Limitations
231
 
 
1
  ---
2
  dataset_info:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  features:
4
  - name: audio
5
  dtype:
 
18
  download_size: 3346820592
19
  dataset_size: 3341105554.6299996
20
  configs:
 
 
 
 
21
  - config_name: default
22
  data_files:
23
  - split: train
 
30
 
31
  This dataset is derived from espnet/yodas, more details can be found here: https://huggingface.co/datasets/espnet/yodas
32
 
33
+ This is a subset of the zh000 subset of espnet/yodas dataset, which selects videos with Mandarin-English code-switching phenomenon.
34
+
35
+ Note that code-switching is only gauranteed per video rather than per utterance. Therefore, not every utterance in the dataset contains code-switching.
36
 
37
  ## Dataset Details
38
 
39
+ ### Dataset Usage
40
+ The `default` config does not modify any text of the selected samples.
41
+ ```python
42
+ from datasets import load_dataset
43
+ cs_yodas = load_dataset("georgechang8/code_switch_yodas_zh")
44
+ ```
45
+ The `clean` config cleanses the text of the selected samples (as in the processing).
46
+ ```python
47
+ from datasets import load_dataset
48
+ cs_yodas_clean = load_dataset("georgechang8/code_switch_yodas_zh", "clean")
49
+ ```
50
+ ```python
51
+ {'audio': {'path': 'GaUSbuZm5Ec-00207-00083809-00084143.wav',
52
+ 'array': array([-0.09082031, 0.01898193, 0.02850342, ..., 0.01419067,
53
+ 0.01391602, 0.01513672]),
54
+ 'sampling_rate': 16000},
55
+ 'text': '項明生,訂Agoda的項明生',
56
+ 'id': 'GaUSbuZm5Ec-00207-00083809-00084143',
57
+ 'session_id': 'GaUSbuZm5Ec'}
58
+ ```
59
+
60
  ### Dataset Description
61
 
62
  <!-- Provide a longer summary of what this dataset is. -->
 
226
  embed_external_files=True
227
  )
228
  ```
229
+ #### Data Cleaning
230
+ 1. The video `Pew9CK74axu` is manually cleaned
231
+ ```python
232
+ def filter_fn(batch):
233
+ return (z == 'Pew9CK74axu' for z in batch['session_id'])
234
+
235
+ special_care = audio_dataset.filter(filter_fn, num_proc=8, batched=True)
236
+ with open("manual_edit.txt", "w", encoding="utf8") as f:
237
+ for l in special_care['text']:
238
+ f.write(l + "\n")
239
+ # manual cleaning ...
240
+ with open("manual_edit_finish.txt", "r", encoding="utf8") as f:
241
+ lines = list(map(str.strip, f.readlines()))
242
+ replace_dict = {
243
+ a: b
244
+ for a, b in zip(special_care['id'], lines)
245
+ }
246
+ def manual_edit(batch):
247
+ texts = []
248
+ for sid, orig in zip(batch['id'], batch['text']):
249
+ texts += [replace_dict.get(sid, orig)]
250
+ return {'text': texts}
251
+
252
+ audio_dataset_manual = audio_dataset.map(manual_edit, batched=True, num_proc=8)
253
+ ```
254
+ 2. General cleansing pipeline
255
+ ```python
256
+ import re
257
+ import html
258
+
259
+ def remove_emojies(text):
260
+ # Ref: https://gist.github.com/Alex-Just/e86110836f3f93fe7932290526529cd1#gistcomment-3208085
261
+ # Ref: https://en.wikipedia.org/wiki/Unicode_block
262
+ EMOJI_PATTERN = re.compile(
263
+ "["
264
+ "\U0001F1E0-\U0001F1FF" # flags (iOS)
265
+ "\U0001F300-\U0001F5FF" # symbols & pictographs
266
+ "\U0001F600-\U0001F64F" # emoticons
267
+ "\U0001F680-\U0001F6FF" # transport & map symbols
268
+ "\U0001F700-\U0001F77F" # alchemical symbols
269
+ "\U0001F780-\U0001F7FF" # Geometric Shapes Extended
270
+ "\U0001F800-\U0001F8FF" # Supplemental Arrows-C
271
+ "\U0001F900-\U0001F9FF" # Supplemental Symbols and Pictographs
272
+ "\U0001FA00-\U0001FA6F" # Chess Symbols
273
+ "\U0001FA70-\U0001FAFF" # Symbols and Pictographs Extended-A
274
+ "\U00002702-\U000027B0" # Dingbats
275
+ "]"
276
+ )
277
+ text = re.sub(EMOJI_PATTERN, r' ', text)
278
+ return text
279
+
280
+ def clean_transcripts(x):
281
+ cjk = "[\u3400-\u4db5\u4e00-\u9fa5\u9fa6-\u9fbb\uf900-\ufa2d\ufa30-\ufa6a\ufa70-\ufad9\uff00-\uffef\u2e80-\u2eff\u3000-\u303f\u31c0-\u31ef\u2f00-\u2fdf\u2ff0-\u2fff\u3100-\u312f\u31a0-\u31bf\ufe10-\ufe1f\ufe30-\ufe4f\u2600-\u26ff\u2700-\u27bf\u3200-\u32ff\u3300-\u33ff]"
282
+ x = html.unescape(x)
283
+ x = remove_emojies(x)
284
+ dots = '\.{3,}'
285
+ x = re.sub(rf'{dots}|…|\s|^|$', ' ', x) # expanding space allows matching " uh uh" case
286
+ x = re.sub(rf"({cjk}|\s)([Uu][mh]|U[MH])({cjk}|\s)", r"\1 \3", x) # uh/um surrounded by cjk or space
287
+ x = re.sub(r"([HhEe]mm+|[HE]MM+)", " ", x) # hmm emm
288
+ x = re.sub(fr"\*+({cjk}+|[A-Za-z]+)\*+", " ", x) # *叹气*
289
+ x = re.sub(r'[呃嗯]', ' ', x) # 呃嗯
290
+ def replace_except(pattern, repl, z, excs):
291
+ for e, t in excs:
292
+ z = z.replace(e, t)
293
+ z = re.sub(pattern, repl, z)
294
+ for e, t in excs:
295
+ z = z.replace(t, e)
296
+ return z
297
+ # remove 恩 except for 恩桥 感恩 恩怨
298
+ x = replace_except("恩", ' ', x, excs=[("感恩", "呃"),("恩桥", "嗯"),("恩怨", "emm")])
299
+ # remove (...) except for 'Program Files (x86)'
300
+ x = re.sub(r'([^()]*)', ' ', x)
301
+ x = re.sub(r"\s+", " ", x)
302
+ x = replace_except(r'\([^()]*\)', ' ', x, excs=[("Program Files (x86)", "呃")])
303
+ puncs = r'[,?!。;?!,;~~]'
304
+ x = re.sub(rf'({puncs})(?:\s*\1)+', r'\1', x) # ??? -> ?
305
+ x = re.sub(rf"\s+({puncs})", r'\1', x) # text , -> text,
306
+ sp_puncs = r'[?!,;]' # puncs with spaces
307
+ x = re.sub(rf"({puncs}*{sp_puncs})([a-zA-Z])", r'\1 \2', x) # text,cont -> text, cont
308
+ x = re.sub(rf"^[\s]*{puncs}+", "", x) # leading puncs
309
+ x = re.sub(r"\s+", " ", x) # excess spaces
310
+ return x.strip()
311
+
312
+ audio_dataset_manual_clean = audio_dataset_manual.map(lambda x: {"text": list(map(clean_transcripts, x['text']))}, batched=True, num_proc=8)
313
+ # push to hub
314
+ audio_dataset_manual_clean.push_to_hub(
315
+ "georgechang8/code_switch_yodas_zh",
316
+ config_name="clean",
317
+ set_default=False,
318
+ commit_message="Clean transcript",
319
+ max_shard_size="1GB",
320
+ embed_external_files=True,
321
+ )
322
+ ```
323
 
324
  ## Limitations
325