RodrigoLimaRFL commited on
Commit
09857e0
·
verified ·
1 Parent(s): f8bbf87

Update NURC-SP_ENTOA_TTS.py

Browse files
Files changed (1) hide show
  1. NURC-SP_ENTOA_TTS.py +67 -156
NURC-SP_ENTOA_TTS.py CHANGED
@@ -2,112 +2,44 @@ import csv
2
  import datasets
3
  from datasets import BuilderConfig, GeneratorBasedBuilder, DatasetInfo, SplitGenerator, Split
4
 
5
- _PROSODIC_PROMPTS_URLS = {
6
- "validation": "prosodic/validation.csv",
7
- "train": "prosodic/train.csv",
 
 
8
  }
9
 
10
- _AUTOMATIC_PROMPTS_URLS = {
11
- "validation": "automatic/validation.csv",
12
  "train": "automatic/train.csv",
13
  }
14
 
15
  _ARCHIVES = {
16
- "prosodic": "prosodic.tar.gz",
17
- "automatic": "automatic.tar.gz",
18
  }
19
 
20
  _PATH_TO_CLIPS = {
21
- "validation_prosodic": "",
22
- "train_prosodic": "",
23
- "validation_automatic": "automatic/validation",
24
- "train_automatic": "automatic/train",
25
  }
26
 
27
- def debug_path_matching(csv_path, archive_files):
28
- """
29
- Debug utility to compare paths between CSV and archive files
30
- """
31
- import csv
32
- from collections import defaultdict
33
-
34
- # Store CSV paths
35
- csv_paths = set()
36
-
37
- with open(csv_path, "r") as f:
38
- reader = csv.DictReader(f)
39
- for row in reader:
40
- # Store both the full path and filename
41
- path = row.get("path") or row.get("file_path")
42
- csv_paths.add(path)
43
- csv_paths.add(path.split("/")[-1])
44
-
45
- # Compare with archive paths
46
- archive_paths = set()
47
- matches = defaultdict(list)
48
-
49
- for path, _ in archive_files:
50
- archive_paths.add(path)
51
- archive_paths.add(path.split("/")[-1])
52
-
53
- # Check for matches
54
- for csv_path in csv_paths:
55
- if path.endswith(csv_path) or csv_path.endswith(path):
56
- matches[path].append(csv_path)
57
-
58
- print("=== Debug Report ===")
59
- print(f"CSV Paths: {len(csv_paths)}")
60
- print(f"Archive Paths: {len(archive_paths)}")
61
- print(f"Matched Paths: {len(matches)}")
62
- print("\nSample CSV paths:")
63
- for path in list(csv_paths)[:5]:
64
- print(f" {path}")
65
- print("\nSample Archive paths:")
66
- for path in list(archive_paths)[:5]:
67
- print(f" {path}")
68
- print("\nSample Matches:")
69
- for archive_path, csv_paths in list(matches.items())[:5]:
70
- print(f" Archive: {archive_path}")
71
- print(f" CSV: {csv_paths}")
72
- print()
73
-
74
- return csv_paths, archive_paths, matches
75
-
76
- class EntoaConfig(BuilderConfig):
77
- def __init__(self, prompts_type="prosodic", **kwargs):
78
  super().__init__(**kwargs)
79
  self.prompts_type = prompts_type
80
 
81
- class EntoaDataset(GeneratorBasedBuilder):
 
82
  BUILDER_CONFIGS = [
83
- EntoaConfig(name="prosodic", description="Prosodic audio prompts", prompts_type="prosodic"),
84
- EntoaConfig(name="automatic", description="Automatic audio prompts", prompts_type="automatic"),
85
  ]
86
 
87
  def _info(self):
88
- if self.config.name == "prosodic":
89
- features = datasets.Features(
90
- {
91
- "path": datasets.Value("string"),
92
- "name": datasets.Value("string"),
93
- "speaker": datasets.Value("string"),
94
- "start_time": datasets.Value("string"),
95
- "end_time": datasets.Value("string"),
96
- "normalized_text": datasets.Value("string"),
97
- "text": datasets.Value("string"),
98
- "duration": datasets.Value("string"),
99
- "type": datasets.Value("string"),
100
- "year": datasets.Value("string"),
101
- "gender": datasets.Value("string"),
102
- "age_range": datasets.Value("string"),
103
- "total_duration": datasets.Value("string"),
104
- "quality": datasets.Value("string"),
105
- "theme": datasets.Value("string"),
106
- "audio": datasets.Audio(sampling_rate=16_000),
107
- }
108
- )
109
- else: # automatic
110
- features = datasets.Features(
111
  {
112
  "audio_name": datasets.Value("string"),
113
  "file_path": datasets.Value("string"),
@@ -127,102 +59,81 @@ class EntoaDataset(GeneratorBasedBuilder):
127
  "audio": datasets.Audio(sampling_rate=16_000),
128
  }
129
  )
130
- return DatasetInfo(features=features)
131
 
132
  def _split_generators(self, dl_manager):
133
- prompts_urls = _PROSODIC_PROMPTS_URLS if self.config.name == "prosodic" else _AUTOMATIC_PROMPTS_URLS
134
- archive = dl_manager.download(_ARCHIVES[self.config.name])
135
- prompts_path = dl_manager.download(prompts_urls)
136
 
137
- # Debug prints for downloaded paths
138
- print(f"Downloaded prompts: {prompts_path}")
139
- print(f"Downloaded archive: {archive}")
 
 
140
 
141
  return [
142
  SplitGenerator(
143
  name=Split.VALIDATION,
144
  gen_kwargs={
145
- "prompts_path": prompts_path["validation"],
146
- "path_to_clips": _PATH_TO_CLIPS[f"validation_{self.config.name}"],
147
- "audio_files": dl_manager.iter_archive(archive),
148
- },
149
  ),
150
  SplitGenerator(
151
  name=Split.TRAIN,
152
  gen_kwargs={
153
  "prompts_path": prompts_path["train"],
154
- "path_to_clips": _PATH_TO_CLIPS[f"train_{self.config.name}"],
155
- "audio_files": dl_manager.iter_archive(archive),
156
- },
157
  ),
158
  ]
159
 
160
-
161
-
162
-
163
  def _generate_examples(self, prompts_path, path_to_clips, audio_files):
164
- csv_paths, archive_paths, matches = debug_path_matching(prompts_path, audio_files)
165
  examples = {}
166
  with open(prompts_path, "r") as f:
167
  csv_reader = csv.DictReader(f)
168
  for row in csv_reader:
169
-
170
- if self.config.name == "prosodic":
171
- examples[row["path"]] = {
172
- "path": row["path"],
173
- "name": row["name"],
174
- "speaker": row["speaker"],
175
- "start_time": row["start_time"],
176
- "end_time": row["end_time"],
177
- "normalized_text": row["normalized_text"],
178
- "text": row["text"],
179
- "duration": row["duration"],
180
- "type": row["type"],
181
- "year": row["year"],
182
- "gender": row["gender"],
183
- "age_range": row["age_range"],
184
- "total_duration": row["total_duration"],
185
- "quality": row["quality"],
186
- "theme": row["theme"],
187
- }
188
- else: # automatic
189
- examples[row["file_path"]] = {
190
- "audio_name": row["audio_name"],
191
- "file_path": row["file_path"],
192
- "text": row["text"],
193
- "start_time": row["start_time"],
194
- "end_time": row["end_time"],
195
- "duration": row["duration"],
196
- "quality": row["quality"],
197
- "speech_genre": row["speech_genre"],
198
- "speech_style": row["speech_style"],
199
- "variety": row["variety"],
200
- "accent": row["accent"],
201
- "sex": row["sex"],
202
- "age_range": row["age_range"],
203
- "num_speakers": row["num_speakers"],
204
- "speaker_id": row["speaker_id"],
205
- }
206
-
207
- id_ = 0
208
  inside_clips_dir = False
209
-
210
  for path, f in audio_files:
211
-
212
  if path.startswith(path_to_clips):
213
  inside_clips_dir = True
214
  if path in examples:
215
- # Debug: Match found
216
- print(f"Match found for: {path}")
217
  audio = {"path": path, "bytes": f.read()}
218
  yield id_, {**examples[path], "audio": audio}
219
  id_ += 1
220
- else:
221
- # Debug: No match for this file
222
- print(f"No match for: {path}")
223
  elif inside_clips_dir:
224
  break
225
-
226
- # Debug: Print total examples generated
227
- print(f"Completed generating examples. Total examples: {id_}")
228
-
 
2
  import datasets
3
  from datasets import BuilderConfig, GeneratorBasedBuilder, DatasetInfo, SplitGenerator, Split
4
 
5
+
6
+
7
+ _PROMPTS_URLS = {
8
+ "dev": "automatic/validation.csv",
9
+ "train": "automatic/train.csv",
10
  }
11
 
12
+ _PROMPTS_FILTERED_URLS = {
13
+ "dev": "automatic/validation.csv",
14
  "train": "automatic/train.csv",
15
  }
16
 
17
  _ARCHIVES = {
18
+ "dev": "automatic.tar.gz",
19
+ "train": "automatic.tar.gz",
20
  }
21
 
22
  _PATH_TO_CLIPS = {
23
+ "dev": "validation",
24
+ "train": "train",
 
 
25
  }
26
 
27
+
28
+ class NurcSPConfig(BuilderConfig):
29
+ def __init__(self, prompts_type="original", **kwargs):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  super().__init__(**kwargs)
31
  self.prompts_type = prompts_type
32
 
33
+
34
+ class NurcSPDataset(GeneratorBasedBuilder):
35
  BUILDER_CONFIGS = [
36
+ NurcSPConfig(name="original", description="Original audio prompts", prompts_type="original"),
37
+ NurcSPConfig(name="filtered", description="Filtered audio prompts", prompts_type="filtered"),
38
  ]
39
 
40
  def _info(self):
41
+ return DatasetInfo(
42
+ features=datasets.Features(
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
43
  {
44
  "audio_name": datasets.Value("string"),
45
  "file_path": datasets.Value("string"),
 
59
  "audio": datasets.Audio(sampling_rate=16_000),
60
  }
61
  )
62
+ )
63
 
64
  def _split_generators(self, dl_manager):
65
+ prompts_urls = _PROMPTS_URLS # Default to original prompts URLs
 
 
66
 
67
+ if self.config.prompts_type == "filtered":
68
+ prompts_urls = _PROMPTS_FILTERED_URLS
69
+
70
+ prompts_path = dl_manager.download(prompts_urls)
71
+ archive = dl_manager.download(_ARCHIVES)
72
 
73
  return [
74
  SplitGenerator(
75
  name=Split.VALIDATION,
76
  gen_kwargs={
77
+ "prompts_path": prompts_path["dev"],
78
+ "path_to_clips": _PATH_TO_CLIPS["dev"],
79
+ "audio_files": dl_manager.iter_archive(archive["dev"]),
80
+ }
81
  ),
82
  SplitGenerator(
83
  name=Split.TRAIN,
84
  gen_kwargs={
85
  "prompts_path": prompts_path["train"],
86
+ "path_to_clips": _PATH_TO_CLIPS["train"],
87
+ "audio_files": dl_manager.iter_archive(archive["train"]),
88
+ }
89
  ),
90
  ]
91
 
 
 
 
92
  def _generate_examples(self, prompts_path, path_to_clips, audio_files):
 
93
  examples = {}
94
  with open(prompts_path, "r") as f:
95
  csv_reader = csv.DictReader(f)
96
  for row in csv_reader:
97
+ audio_name = row['audio_name']
98
+ file_path = row['file_path']
99
+ text = row['text']
100
+ start_time = row['start_time']
101
+ end_time = row['end_time']
102
+ duration = row['duration']
103
+ quality = row['quality']
104
+ speech_genre = row['speech_genre']
105
+ speech_style = row['speech_style']
106
+ variety = row['variety']
107
+ accent = row['accent']
108
+ sex = row['sex']
109
+ age_range = row['age_range']
110
+ num_speakers = row['num_speakers']
111
+ speaker_id = row['speaker_id']
112
+ examples[file_path] = {
113
+ "audio_name": audio_name,
114
+ "file_path": file_path,
115
+ "text": text,
116
+ "start_time": start_time,
117
+ "end_time": end_time,
118
+ "duration": duration,
119
+ "quality": quality,
120
+ "speech_genre": speech_genre,
121
+ "speech_style": speech_style,
122
+ "variety": variety,
123
+ "accent": accent,
124
+ "sex": sex,
125
+ "age_range": age_range,
126
+ "num_speakers": num_speakers,
127
+ "speaker_id": speaker_id,
128
+ }
 
 
 
 
 
 
 
129
  inside_clips_dir = False
130
+ id_ = 0
131
  for path, f in audio_files:
 
132
  if path.startswith(path_to_clips):
133
  inside_clips_dir = True
134
  if path in examples:
 
 
135
  audio = {"path": path, "bytes": f.read()}
136
  yield id_, {**examples[path], "audio": audio}
137
  id_ += 1
 
 
 
138
  elif inside_clips_dir:
139
  break