Datasets:
Languages:
English
ArXiv:
Tags:
query-by-example-spoken-term-detection
audio-slot-filling
speaker-diarization
automatic-speaker-verification
License:
Update files from the datasets library (from 1.14.0)
Browse filesRelease notes: https://github.com/huggingface/datasets/releases/tag/1.14.0
README.md
CHANGED
@@ -73,7 +73,7 @@ SUPERB is a leaderboard to benchmark the performance of a shared model across a
|
|
73 |
|
74 |
### Supported Tasks and Leaderboards
|
75 |
|
76 |
-
The SUPERB leaderboard can be found here
|
77 |
|
78 |
#### pr
|
79 |
|
@@ -205,6 +205,9 @@ An example from each split looks like:
|
|
205 |
```python
|
206 |
{'chapter_id': 1240,
|
207 |
'file': 'path/to/file.flac',
|
|
|
|
|
|
|
208 |
'id': '103-1240-0000',
|
209 |
'speaker_id': 103,
|
210 |
'text': 'CHAPTER ONE MISSUS RACHEL LYNDE IS SURPRISED MISSUS RACHEL LYNDE '
|
@@ -220,6 +223,9 @@ An example from each split looks like:
|
|
220 |
```python
|
221 |
{
|
222 |
'file': '/path/yes/af7a8296_nohash_1.wav',
|
|
|
|
|
|
|
223 |
'label': 0 # 'yes'
|
224 |
}
|
225 |
```
|
@@ -234,6 +240,9 @@ An example from each split looks like:
|
|
234 |
```python
|
235 |
{
|
236 |
'file': "/path/wavs/speakers/2BqVo8kVB2Skwgyb/063aa8f0-4479-11e9-a9a5-5dbec3b8816a.wav",
|
|
|
|
|
|
|
237 |
'speaker_id': '2BqVo8kVB2Skwgyb',
|
238 |
'text': 'Turn the bedroom lights off',
|
239 |
'action': 3, # 'deactivate'
|
@@ -252,6 +261,9 @@ An example from each split looks like:
|
|
252 |
```python
|
253 |
{
|
254 |
'file': '/path/wav/id10003/na8-QEFmj44/00003.wav',
|
|
|
|
|
|
|
255 |
'label': 2 # 'id10003'
|
256 |
}
|
257 |
```
|
@@ -268,6 +280,9 @@ An example from each split looks like:
|
|
268 |
{
|
269 |
'record_id': '1578-6379-0038_6415-111615-0009',
|
270 |
'file': 'path/to/file.wav',
|
|
|
|
|
|
|
271 |
'start': 0,
|
272 |
'end': 1590,
|
273 |
'speakers': [
|
@@ -287,6 +302,10 @@ An example from each split looks like:
|
|
287 |
|
288 |
### Data Fields
|
289 |
|
|
|
|
|
|
|
|
|
290 |
#### pr
|
291 |
|
292 |
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
@@ -294,15 +313,17 @@ An example from each split looks like:
|
|
294 |
|
295 |
#### asr
|
296 |
|
297 |
-
- `file
|
298 |
-
- `
|
299 |
-
- `
|
300 |
-
- `
|
301 |
-
- `
|
|
|
302 |
|
303 |
#### ks
|
304 |
|
305 |
- `file` (`string`): Path to the WAV audio file.
|
|
|
306 |
- `label` (`ClassLabel`): Label of the spoken command. Possible values:
|
307 |
- `0: "yes", 1: "no", 2: "up", 3: "down", 4: "left", 5: "right", 6: "on", 7: "off", 8: "stop", 9: "go", 10: "_silence_", 11: "_unknown_"`
|
308 |
|
@@ -313,6 +334,7 @@ An example from each split looks like:
|
|
313 |
#### ic
|
314 |
|
315 |
- `file` (`string`): Path to the WAV audio file.
|
|
|
316 |
- `speaker_id` (`string`): ID of the speaker.
|
317 |
- `text` (`string`): Transcription of the spoken command.
|
318 |
- `action` (`ClassLabel`): Label of the command's action. Possible values:
|
@@ -330,6 +352,7 @@ An example from each split looks like:
|
|
330 |
#### si
|
331 |
|
332 |
- `file` (`string`): Path to the WAV audio file.
|
|
|
333 |
- `label` (`ClassLabel`): Label (ID) of the speaker. Possible values:
|
334 |
- `0: "id10001", 1: "id10002", 2: "id10003", ..., 1250: "id11251"`
|
335 |
|
@@ -343,6 +366,7 @@ An example from each split looks like:
|
|
343 |
The data fields in all splits are:
|
344 |
- `record_id` (`string`): ID of the record.
|
345 |
- `file` (`string`): Path to the WAV audio file.
|
|
|
346 |
- `start` (`integer`): Start frame of the audio.
|
347 |
- `end` (`integer`): End frame of the audio.
|
348 |
- `speakers` (`list` of `dict`): List of speakers in the audio. Each item contains the fields:
|
@@ -353,6 +377,7 @@ The data fields in all splits are:
|
|
353 |
#### er
|
354 |
|
355 |
- `file` (`string`): Path to the WAV audio file.
|
|
|
356 |
- `label` (`ClassLabel`): Label of the speech emotion. Possible values:
|
357 |
- `0: "neu", 1: "hap", 2: "ang", 3: "sad"`
|
358 |
|
|
|
73 |
|
74 |
### Supported Tasks and Leaderboards
|
75 |
|
76 |
+
The SUPERB leaderboard can be found here https://superbbenchmark.org/leaderboard and consists of the following tasks:
|
77 |
|
78 |
#### pr
|
79 |
|
|
|
205 |
```python
|
206 |
{'chapter_id': 1240,
|
207 |
'file': 'path/to/file.flac',
|
208 |
+
'audio': {'path': 'path/to/file.flac',
|
209 |
+
'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
|
210 |
+
'sampling_rate': 16000},
|
211 |
'id': '103-1240-0000',
|
212 |
'speaker_id': 103,
|
213 |
'text': 'CHAPTER ONE MISSUS RACHEL LYNDE IS SURPRISED MISSUS RACHEL LYNDE '
|
|
|
223 |
```python
|
224 |
{
|
225 |
'file': '/path/yes/af7a8296_nohash_1.wav',
|
226 |
+
'audio': {'path': '/path/yes/af7a8296_nohash_1.wav',
|
227 |
+
'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
|
228 |
+
'sampling_rate': 16000},
|
229 |
'label': 0 # 'yes'
|
230 |
}
|
231 |
```
|
|
|
240 |
```python
|
241 |
{
|
242 |
'file': "/path/wavs/speakers/2BqVo8kVB2Skwgyb/063aa8f0-4479-11e9-a9a5-5dbec3b8816a.wav",
|
243 |
+
'audio': {'path': '/path/wavs/speakers/2BqVo8kVB2Skwgyb/063aa8f0-4479-11e9-a9a5-5dbec3b8816a.wav',
|
244 |
+
'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
|
245 |
+
'sampling_rate': 16000},
|
246 |
'speaker_id': '2BqVo8kVB2Skwgyb',
|
247 |
'text': 'Turn the bedroom lights off',
|
248 |
'action': 3, # 'deactivate'
|
|
|
261 |
```python
|
262 |
{
|
263 |
'file': '/path/wav/id10003/na8-QEFmj44/00003.wav',
|
264 |
+
'audio': {'path': '/path/wav/id10003/na8-QEFmj44/00003.wav',
|
265 |
+
'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
|
266 |
+
'sampling_rate': 16000},
|
267 |
'label': 2 # 'id10003'
|
268 |
}
|
269 |
```
|
|
|
280 |
{
|
281 |
'record_id': '1578-6379-0038_6415-111615-0009',
|
282 |
'file': 'path/to/file.wav',
|
283 |
+
'audio': {'path': 'path/to/file.wav',
|
284 |
+
'array': array([-0.00048828, -0.00018311, -0.00137329, ..., 0.00079346, 0.00091553, 0.00085449], dtype=float32),
|
285 |
+
'sampling_rate': 16000},
|
286 |
'start': 0,
|
287 |
'end': 1590,
|
288 |
'speakers': [
|
|
|
302 |
|
303 |
### Data Fields
|
304 |
|
305 |
+
####Note abouth the `audio` fields
|
306 |
+
|
307 |
+
When accessing the audio column: `dataset[0]["audio"]` the audio file is automatically decoded and resampled to `dataset.features["audio"].sampling_rate`. Decoding and resampling of a large number of audio files might take a significant amount of time. Thus it is important to first query the sample index before the `"audio"` column, *i.e.* `dataset[0]["audio"]` should **always** be preferred over `dataset["audio"][0]`.
|
308 |
+
|
309 |
#### pr
|
310 |
|
311 |
[More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
|
|
313 |
|
314 |
#### asr
|
315 |
|
316 |
+
- `file` (`string`): Path to the WAV audio file.
|
317 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
318 |
+
- `text` (`string`): The transcription of the audio file.
|
319 |
+
- `speaker_id` (`integer`): A unique ID of the speaker. The same speaker id can be found for multiple data samples.
|
320 |
+
- `chapter_id` (`integer`): ID of the audiobook chapter which includes the transcription.
|
321 |
+
- `id` (`string`): A unique ID of the data sample.
|
322 |
|
323 |
#### ks
|
324 |
|
325 |
- `file` (`string`): Path to the WAV audio file.
|
326 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
327 |
- `label` (`ClassLabel`): Label of the spoken command. Possible values:
|
328 |
- `0: "yes", 1: "no", 2: "up", 3: "down", 4: "left", 5: "right", 6: "on", 7: "off", 8: "stop", 9: "go", 10: "_silence_", 11: "_unknown_"`
|
329 |
|
|
|
334 |
#### ic
|
335 |
|
336 |
- `file` (`string`): Path to the WAV audio file.
|
337 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
338 |
- `speaker_id` (`string`): ID of the speaker.
|
339 |
- `text` (`string`): Transcription of the spoken command.
|
340 |
- `action` (`ClassLabel`): Label of the command's action. Possible values:
|
|
|
352 |
#### si
|
353 |
|
354 |
- `file` (`string`): Path to the WAV audio file.
|
355 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
356 |
- `label` (`ClassLabel`): Label (ID) of the speaker. Possible values:
|
357 |
- `0: "id10001", 1: "id10002", 2: "id10003", ..., 1250: "id11251"`
|
358 |
|
|
|
366 |
The data fields in all splits are:
|
367 |
- `record_id` (`string`): ID of the record.
|
368 |
- `file` (`string`): Path to the WAV audio file.
|
369 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
370 |
- `start` (`integer`): Start frame of the audio.
|
371 |
- `end` (`integer`): End frame of the audio.
|
372 |
- `speakers` (`list` of `dict`): List of speakers in the audio. Each item contains the fields:
|
|
|
377 |
#### er
|
378 |
|
379 |
- `file` (`string`): Path to the WAV audio file.
|
380 |
+
- `audio` (`dict`): A dictionary containing the path to the downloaded audio file, the decoded audio array, and the sampling rate.
|
381 |
- `label` (`ClassLabel`): Label of the speech emotion. Possible values:
|
382 |
- `0: "neu", 1: "hap", 2: "ang", 3: "sad"`
|
383 |
|
superb.py
CHANGED
@@ -137,6 +137,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
137 |
features=datasets.Features(
|
138 |
{
|
139 |
"file": datasets.Value("string"),
|
|
|
140 |
"text": datasets.Value("string"),
|
141 |
"speaker_id": datasets.Value("int64"),
|
142 |
"chapter_id": datasets.Value("int64"),
|
@@ -161,6 +162,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
161 |
features=datasets.Features(
|
162 |
{
|
163 |
"file": datasets.Value("string"),
|
|
|
164 |
"label": datasets.ClassLabel(
|
165 |
names=[
|
166 |
"yes",
|
@@ -194,6 +196,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
194 |
features=datasets.Features(
|
195 |
{
|
196 |
"file": datasets.Value("string"),
|
|
|
197 |
"speaker_id": datasets.Value("string"),
|
198 |
"text": datasets.Value("string"),
|
199 |
"action": datasets.ClassLabel(
|
@@ -235,6 +238,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
235 |
features=datasets.Features(
|
236 |
{
|
237 |
"file": datasets.Value("string"),
|
|
|
238 |
# VoxCeleb1 contains 1251 speaker IDs in range ["id10001",..."id11251"]
|
239 |
"label": datasets.ClassLabel(names=[f"id{i + 10001}" for i in range(1251)]),
|
240 |
}
|
@@ -257,6 +261,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
257 |
{
|
258 |
"record_id": datasets.Value("string"),
|
259 |
"file": datasets.Value("string"),
|
|
|
260 |
"start": datasets.Value("int64"),
|
261 |
"end": datasets.Value("int64"),
|
262 |
"speakers": [
|
@@ -284,6 +289,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
284 |
features=datasets.Features(
|
285 |
{
|
286 |
"file": datasets.Value("string"),
|
|
|
287 |
"label": datasets.ClassLabel(names=["neu", "hap", "ang", "sad"]),
|
288 |
}
|
289 |
),
|
@@ -440,11 +446,13 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
440 |
id_, transcript = line.split(" ", 1)
|
441 |
audio_file = f"{id_}.flac"
|
442 |
speaker_id, chapter_id = [int(el) for el in id_.split("-")[:2]]
|
|
|
443 |
yield key, {
|
444 |
"id": id_,
|
445 |
"speaker_id": speaker_id,
|
446 |
"chapter_id": chapter_id,
|
447 |
-
"file":
|
|
|
448 |
"text": transcript,
|
449 |
}
|
450 |
key += 1
|
@@ -460,7 +468,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
460 |
label = "_silence_"
|
461 |
else:
|
462 |
label = "_unknown_"
|
463 |
-
yield key, {"file": audio_file, "label": label}
|
464 |
elif self.config.name == "ic":
|
465 |
root_path = os.path.join(archive_path, "fluent_speech_commands_dataset")
|
466 |
csv_path = os.path.join(root_path, "data", f"{split}_data.csv")
|
@@ -469,8 +477,10 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
469 |
next(csv_reader)
|
470 |
for row in csv_reader:
|
471 |
key, file_path, speaker_id, text, action, object_, location = row
|
|
|
472 |
yield key, {
|
473 |
-
"file":
|
|
|
474 |
"speaker_id": speaker_id,
|
475 |
"text": text,
|
476 |
"action": action,
|
@@ -486,8 +496,10 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
486 |
if int(split_id) != split:
|
487 |
continue
|
488 |
speaker_id = file_path.split("/")[0]
|
|
|
489 |
yield key, {
|
490 |
-
"file":
|
|
|
491 |
"label": speaker_id,
|
492 |
}
|
493 |
elif self.config.name == "sd":
|
@@ -500,6 +512,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
500 |
yield key, {
|
501 |
"record_id": rec,
|
502 |
"file": data.wavs[rec],
|
|
|
503 |
"start": st,
|
504 |
"end": ed,
|
505 |
"speakers": speakers,
|
@@ -512,6 +525,7 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
512 |
yield key, {
|
513 |
"record_id": rec,
|
514 |
"file": data.wavs[rec],
|
|
|
515 |
"start": st,
|
516 |
"end": ed,
|
517 |
"speakers": speakers,
|
@@ -533,8 +547,10 @@ class Superb(datasets.GeneratorBasedBuilder):
|
|
533 |
continue
|
534 |
wav_subdir = filename.rsplit("_", 1)[0]
|
535 |
filename = f"{filename}.wav"
|
|
|
536 |
yield key, {
|
537 |
-
"file":
|
|
|
538 |
"label": emo.replace("exc", "hap"),
|
539 |
}
|
540 |
key += 1
|
|
|
137 |
features=datasets.Features(
|
138 |
{
|
139 |
"file": datasets.Value("string"),
|
140 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
141 |
"text": datasets.Value("string"),
|
142 |
"speaker_id": datasets.Value("int64"),
|
143 |
"chapter_id": datasets.Value("int64"),
|
|
|
162 |
features=datasets.Features(
|
163 |
{
|
164 |
"file": datasets.Value("string"),
|
165 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
166 |
"label": datasets.ClassLabel(
|
167 |
names=[
|
168 |
"yes",
|
|
|
196 |
features=datasets.Features(
|
197 |
{
|
198 |
"file": datasets.Value("string"),
|
199 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
200 |
"speaker_id": datasets.Value("string"),
|
201 |
"text": datasets.Value("string"),
|
202 |
"action": datasets.ClassLabel(
|
|
|
238 |
features=datasets.Features(
|
239 |
{
|
240 |
"file": datasets.Value("string"),
|
241 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
242 |
# VoxCeleb1 contains 1251 speaker IDs in range ["id10001",..."id11251"]
|
243 |
"label": datasets.ClassLabel(names=[f"id{i + 10001}" for i in range(1251)]),
|
244 |
}
|
|
|
261 |
{
|
262 |
"record_id": datasets.Value("string"),
|
263 |
"file": datasets.Value("string"),
|
264 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
265 |
"start": datasets.Value("int64"),
|
266 |
"end": datasets.Value("int64"),
|
267 |
"speakers": [
|
|
|
289 |
features=datasets.Features(
|
290 |
{
|
291 |
"file": datasets.Value("string"),
|
292 |
+
"audio": datasets.features.Audio(sampling_rate=16_000),
|
293 |
"label": datasets.ClassLabel(names=["neu", "hap", "ang", "sad"]),
|
294 |
}
|
295 |
),
|
|
|
446 |
id_, transcript = line.split(" ", 1)
|
447 |
audio_file = f"{id_}.flac"
|
448 |
speaker_id, chapter_id = [int(el) for el in id_.split("-")[:2]]
|
449 |
+
audio_path = os.path.join(transcript_dir_path, audio_file)
|
450 |
yield key, {
|
451 |
"id": id_,
|
452 |
"speaker_id": speaker_id,
|
453 |
"chapter_id": chapter_id,
|
454 |
+
"file": audio_path,
|
455 |
+
"audio": audio_path,
|
456 |
"text": transcript,
|
457 |
}
|
458 |
key += 1
|
|
|
468 |
label = "_silence_"
|
469 |
else:
|
470 |
label = "_unknown_"
|
471 |
+
yield key, {"file": audio_file, "audio": audio_file, "label": label}
|
472 |
elif self.config.name == "ic":
|
473 |
root_path = os.path.join(archive_path, "fluent_speech_commands_dataset")
|
474 |
csv_path = os.path.join(root_path, "data", f"{split}_data.csv")
|
|
|
477 |
next(csv_reader)
|
478 |
for row in csv_reader:
|
479 |
key, file_path, speaker_id, text, action, object_, location = row
|
480 |
+
audio_path = os.path.join(root_path, file_path)
|
481 |
yield key, {
|
482 |
+
"file": audio_path,
|
483 |
+
"audio": audio_path,
|
484 |
"speaker_id": speaker_id,
|
485 |
"text": text,
|
486 |
"action": action,
|
|
|
496 |
if int(split_id) != split:
|
497 |
continue
|
498 |
speaker_id = file_path.split("/")[0]
|
499 |
+
audio_path = os.path.join(wav_path, file_path)
|
500 |
yield key, {
|
501 |
+
"file": audio_path,
|
502 |
+
"audio": audio_path,
|
503 |
"label": speaker_id,
|
504 |
}
|
505 |
elif self.config.name == "sd":
|
|
|
512 |
yield key, {
|
513 |
"record_id": rec,
|
514 |
"file": data.wavs[rec],
|
515 |
+
"audio": data.wavs[rec],
|
516 |
"start": st,
|
517 |
"end": ed,
|
518 |
"speakers": speakers,
|
|
|
525 |
yield key, {
|
526 |
"record_id": rec,
|
527 |
"file": data.wavs[rec],
|
528 |
+
"audio": data.wavs[rec],
|
529 |
"start": st,
|
530 |
"end": ed,
|
531 |
"speakers": speakers,
|
|
|
547 |
continue
|
548 |
wav_subdir = filename.rsplit("_", 1)[0]
|
549 |
filename = f"{filename}.wav"
|
550 |
+
audio_path = os.path.join(wav_path, wav_subdir, filename)
|
551 |
yield key, {
|
552 |
+
"file": audio_path,
|
553 |
+
"audio": audio_path,
|
554 |
"label": emo.replace("exc", "hap"),
|
555 |
}
|
556 |
key += 1
|