subatomicseer
commited on
Commit
•
536046d
1
Parent(s):
3096337
Create speech_segment.py
Browse files- speech_segment.py +101 -0
speech_segment.py
ADDED
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Lint as: python3
|
2 |
+
"""Speech Segment dataset.
|
3 |
+
"""
|
4 |
+
import os
|
5 |
+
from pathlib import Path
|
6 |
+
|
7 |
+
import datasets
|
8 |
+
import torchaudio
|
9 |
+
|
10 |
+
|
11 |
+
class SpeechSegmentConfig(datasets.BuilderConfig):
|
12 |
+
"""BuilderConfig for Speech Segment.
|
13 |
+
For long audio files, segment them into smaller segments of fixed length.
|
14 |
+
For short audio files, return the whole audio file.
|
15 |
+
"""
|
16 |
+
|
17 |
+
def __init__(self, segment_length, **kwargs):
|
18 |
+
super(SpeechSegmentConfig, self).__init__(**kwargs)
|
19 |
+
self.segment_length = segment_length
|
20 |
+
|
21 |
+
|
22 |
+
class SpeechSegment(datasets.GeneratorBasedBuilder):
|
23 |
+
"""Speech Segment dataset."""
|
24 |
+
|
25 |
+
BUILDER_CONFIGS = [
|
26 |
+
SpeechSegmentConfig(name="all", segment_length=60.0,),
|
27 |
+
]
|
28 |
+
|
29 |
+
@property
|
30 |
+
def manual_download_instructions(self):
|
31 |
+
return (
|
32 |
+
"Specify the data_dir as the path to the folder, will recursively search for .flac and .wav files. "
|
33 |
+
"`datasets.load_dataset('subatomicseer/speech_segment', data_dir='path/to/folder/folder_name')`"
|
34 |
+
)
|
35 |
+
|
36 |
+
def _info(self):
|
37 |
+
features = datasets.Features(
|
38 |
+
{
|
39 |
+
"id": datasets.Value("string"),
|
40 |
+
"file": datasets.Value("string"),
|
41 |
+
'sample_rate': datasets.Value('int64'),
|
42 |
+
'offset': datasets.Value('int64'),
|
43 |
+
'num_frames': datasets.Value('int64'),
|
44 |
+
}
|
45 |
+
)
|
46 |
+
|
47 |
+
return datasets.DatasetInfo(
|
48 |
+
features=features,
|
49 |
+
)
|
50 |
+
|
51 |
+
def _split_generators(self, dl_manager):
|
52 |
+
base_data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
|
53 |
+
if not os.path.exists(base_data_dir):
|
54 |
+
raise FileNotFoundError(
|
55 |
+
f"{base_data_dir} does not exist. Manual download instructions: {self.manual_download_instructions}"
|
56 |
+
)
|
57 |
+
|
58 |
+
data_dirs = [str(p) for p in Path(base_data_dir).rglob('*') if p.suffix in ['.flac', '.wav']]
|
59 |
+
print(f"Found {len(data_dirs)} audio files in {base_data_dir}")
|
60 |
+
return [
|
61 |
+
datasets.SplitGenerator(
|
62 |
+
name=datasets.Split.TRAIN,
|
63 |
+
gen_kwargs={"data_dirs": data_dirs},
|
64 |
+
),
|
65 |
+
]
|
66 |
+
|
67 |
+
def _generate_examples(self, data_dirs):
|
68 |
+
for key, path in enumerate(data_dirs):
|
69 |
+
path_split = path.split("/")
|
70 |
+
id_ = '/'.join(path_split[-4:]).replace(".flac", "")
|
71 |
+
|
72 |
+
audio_metadata = torchaudio.info(path)
|
73 |
+
segment_length = int(self.config.segment_length * audio_metadata.sample_rate)
|
74 |
+
total_length = audio_metadata.num_frames
|
75 |
+
|
76 |
+
if total_length <= segment_length:
|
77 |
+
yield id_, {
|
78 |
+
"id": id_,
|
79 |
+
"file": path,
|
80 |
+
'sample_rate': audio_metadata.sample_rate,
|
81 |
+
'offset': 0,
|
82 |
+
'num_frames': total_length,
|
83 |
+
}
|
84 |
+
else:
|
85 |
+
# generate non-overlapping segments of segment_length
|
86 |
+
offsets = list(range(0, total_length, segment_length))
|
87 |
+
if total_length - offsets[-1] < 1 * audio_metadata.sample_rate:
|
88 |
+
# if the last segment is less than 2 seconds, discard it
|
89 |
+
offsets.pop()
|
90 |
+
|
91 |
+
for segment_id, start in enumerate(offsets):
|
92 |
+
end = start + segment_length - 1
|
93 |
+
if end > total_length:
|
94 |
+
end = total_length
|
95 |
+
yield f'{id_}_{segment_id}', {
|
96 |
+
"id": f'{id_}_{segment_id}',
|
97 |
+
"file": path,
|
98 |
+
'sample_rate': audio_metadata.sample_rate,
|
99 |
+
'offset': start,
|
100 |
+
'num_frames': end-start+1,
|
101 |
+
}
|