system HF staff commited on
Commit
146ebe6
0 Parent(s):

Update files from the datasets library (from 1.13.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.13.0

.gitattributes ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bin.* filter=lfs diff=lfs merge=lfs -text
5
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.model filter=lfs diff=lfs merge=lfs -text
12
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
13
+ *.onnx filter=lfs diff=lfs merge=lfs -text
14
+ *.ot filter=lfs diff=lfs merge=lfs -text
15
+ *.parquet filter=lfs diff=lfs merge=lfs -text
16
+ *.pb filter=lfs diff=lfs merge=lfs -text
17
+ *.pt filter=lfs diff=lfs merge=lfs -text
18
+ *.pth filter=lfs diff=lfs merge=lfs -text
19
+ *.rar filter=lfs diff=lfs merge=lfs -text
20
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
21
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
22
+ *.tflite filter=lfs diff=lfs merge=lfs -text
23
+ *.tgz filter=lfs diff=lfs merge=lfs -text
24
+ *.xz filter=lfs diff=lfs merge=lfs -text
25
+ *.zip filter=lfs diff=lfs merge=lfs -text
26
+ *.zstandard filter=lfs diff=lfs merge=lfs -text
27
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: AMI Corpus
3
+ annotations_creators:
4
+ - expert-generated
5
+ language_creators:
6
+ - crowdsourced
7
+ - expert-generated
8
+ languages:
9
+ - en
10
+ licenses:
11
+ - cc-by-4-0
12
+ multilinguality:
13
+ - monolingual
14
+ size_categories:
15
+ - 100K<n<1M
16
+ source_datasets:
17
+ - original
18
+ task_categories:
19
+ - speech-processing
20
+ task_ids:
21
+ - automatic-speech-recognition
22
+ ---
23
+
24
+ # Dataset Card for AMI Corpus
25
+
26
+ ## Table of Contents
27
+ - [Dataset Description](#dataset-description)
28
+ - [Dataset Summary](#dataset-summary)
29
+ - [Dataset Preprocessing](#dataset-preprocessing)
30
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
31
+ - [Languages](#languages)
32
+ - [Dataset Structure](#dataset-structure)
33
+ - [Data Instances](#data-instances)
34
+ - [Data Fields](#data-fields)
35
+ - [Data Splits](#data-splits)
36
+ - [Dataset Creation](#dataset-creation)
37
+ - [Curation Rationale](#curation-rationale)
38
+ - [Source Data](#source-data)
39
+ - [Annotations](#annotations)
40
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
41
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
42
+ - [Social Impact of Dataset](#social-impact-of-dataset)
43
+ - [Discussion of Biases](#discussion-of-biases)
44
+ - [Other Known Limitations](#other-known-limitations)
45
+ - [Additional Information](#additional-information)
46
+ - [Dataset Curators](#dataset-curators)
47
+ - [Licensing Information](#licensing-information)
48
+ - [Citation Information](#citation-information)
49
+ - [Contributions](#contributions)
50
+
51
+ ## Dataset Description
52
+
53
+ - **Homepage:** [AMI corpus](https://groups.inf.ed.ac.uk/ami/corpus/)
54
+ - **Repository:** [Needs More Information]
55
+ - **Paper:** [Needs More Information]
56
+ - **Leaderboard:** [Needs More Information]
57
+ - **Point of Contact:** [Needs More Information]
58
+
59
+ ### Dataset Summary
60
+
61
+ The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
62
+ synchronized to a common timeline. These include close-talking and far-field microphones, individual and
63
+ room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
64
+ the participants also have unsynchronized pens available to them that record what is written. The meetings
65
+ were recorded in English using three different rooms with different acoustic properties, and include mostly
66
+ non-native speakers.
67
+
68
+ ### Dataset Preprocessing
69
+
70
+ Individual samples of the AMI dataset contain very large audio files (between 10 and 60 minutes).
71
+ Such lengths are unfeasible for most speech recognition models. In the following, we show how the
72
+ dataset can effectively be chunked into multiple segments as defined by the dataset creators.
73
+
74
+ The following function cuts the long audio files into the defined segment lengths:
75
+
76
+ ```python
77
+ import librosa
78
+ import math
79
+ from datasets import load_dataset
80
+
81
+ SAMPLE_RATE = 16_000
82
+
83
+ def chunk_audio(batch):
84
+ new_batch = {
85
+ "audio": [],
86
+ "words": [],
87
+ "speaker": [],
88
+ "lengths": [],
89
+ "word_start_times": [],
90
+ "segment_start_times": [],
91
+ }
92
+
93
+ audio, _ = librosa.load(batch["file"][0], sr=SAMPLE_RATE)
94
+
95
+ word_idx = 0
96
+ num_words = len(batch["words"][0])
97
+ for segment_idx in range(len(batch["segment_start_times"][0])):
98
+ words = []
99
+ word_start_times = []
100
+ start_time = batch["segment_start_times"][0][segment_idx]
101
+ end_time = batch["segment_end_times"][0][segment_idx]
102
+
103
+ # go back and forth with word_idx since segments overlap with each other
104
+ while (word_idx > 1) and (start_time < batch["word_end_times"][0][word_idx - 1]):
105
+ word_idx -= 1
106
+
107
+ while word_idx < num_words and (start_time > batch["word_start_times"][0][word_idx]):
108
+ word_idx += 1
109
+
110
+ new_batch["audio"].append(audio[int(start_time * SAMPLE_RATE): int(end_time * SAMPLE_RATE)])
111
+
112
+ while word_idx < num_words and batch["word_start_times"][0][word_idx] < end_time:
113
+ words.append(batch["words"][0][word_idx])
114
+ word_start_times.append(batch["word_start_times"][0][word_idx])
115
+ word_idx += 1
116
+
117
+ new_batch["lengths"].append(end_time - start_time)
118
+ new_batch["words"].append(words)
119
+ new_batch["speaker"].append(batch["segment_speakers"][0][segment_idx])
120
+ new_batch["word_start_times"].append(word_start_times)
121
+
122
+ new_batch["segment_start_times"].append(batch["segment_start_times"][0][segment_idx])
123
+
124
+ return new_batch
125
+
126
+ ami = load_dataset("ami", "headset-single")
127
+ ami = ami.map(chunk_audio, batched=True, batch_size=1, remove_columns=ami["train"].column_names)
128
+ ```
129
+
130
+ The segmented audio files can still be as long as a minute. To further chunk the data into shorter
131
+ audio chunks, you can use the following script.
132
+
133
+ ```python
134
+ MAX_LENGTH_IN_SECONDS = 20.0
135
+
136
+ def chunk_into_max_n_seconds(batch):
137
+ new_batch = {
138
+ "audio": [],
139
+ "text": [],
140
+ }
141
+
142
+ sample_length = batch["lengths"][0]
143
+ segment_start = batch["segment_start_times"][0]
144
+
145
+ if sample_length > MAX_LENGTH_IN_SECONDS:
146
+ num_chunks_per_sample = math.ceil(sample_length / MAX_LENGTH_IN_SECONDS)
147
+ avg_chunk_length = sample_length / num_chunks_per_sample
148
+ num_words = len(batch["words"][0])
149
+
150
+ # start chunking by times
151
+ start_word_idx = end_word_idx = 0
152
+ chunk_start_time = 0
153
+ for n in range(num_chunks_per_sample):
154
+ while (end_word_idx < num_words - 1) and (batch["word_start_times"][0][end_word_idx] < segment_start + (n + 1) * avg_chunk_length):
155
+ end_word_idx += 1
156
+
157
+ chunk_end_time = int((batch["word_start_times"][0][end_word_idx] - segment_start) * SAMPLE_RATE)
158
+ new_batch["audio"].append(batch["audio"][0][chunk_start_time: chunk_end_time])
159
+ new_batch["text"].append(" ".join(batch["words"][0][start_word_idx: end_word_idx]))
160
+
161
+ chunk_start_time = chunk_end_time
162
+ start_word_idx = end_word_idx
163
+ else:
164
+ new_batch["audio"].append(batch["audio"][0])
165
+ new_batch["text"].append(" ".join(batch["words"][0]))
166
+
167
+ return new_batch
168
+
169
+ ami = ami.map(chunk_into_max_n_seconds, batched=True, batch_size=1, remove_columns=ami["train"].column_names, num_proc=64)
170
+ ```
171
+
172
+ A segmented and chunked dataset of the config `"headset-single"`can be found [here](https://huggingface.co/datasets/ami-wav2vec2/ami_single_headset_segmented_and_chunked).
173
+
174
+ ### Supported Tasks and Leaderboards
175
+
176
+ - `automatic-speech-recognition`: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task does not have an active leaderboard at the moment.
177
+
178
+ - `speaker-diarization`: The dataset can be used to train model for Speaker Diarization (SD). The model is presented with an audio file and asked to predict which speaker spoke at what time.
179
+
180
+ ### Languages
181
+
182
+ The audio is in English.
183
+
184
+ ## Dataset Structure
185
+
186
+ ### Data Instances
187
+
188
+ A typical data point comprises the path to the audio file (or files in the case of
189
+ the multi-headset or multi-microphone dataset), called `file` and its transcription as
190
+ a list of words, called `words`. Additional information about the `speakers`, the `word_start_time`, `word_end_time`, `segment_start_time`, `segment_end_time` is given.
191
+ In addition
192
+
193
+ and its transcription, called `text`. Some additional information about the speaker and the passage which contains the transcription is provided.
194
+
195
+ ```
196
+ {'word_ids': ["ES2004a.D.words1", "ES2004a.D.words2", ...],
197
+ 'word_start_times': [0.3700000047683716, 0.949999988079071, ...],
198
+ 'word_end_times': [0.949999988079071, 1.5299999713897705, ...],
199
+ 'word_speakers': ['A', 'A', ...],
200
+ 'segment_ids': ["ES2004a.sync.1", "ES2004a.sync.2", ...]
201
+ 'segment_start_times': [10.944000244140625, 17.618999481201172, ...],
202
+ 'segment_end_times': [17.618999481201172, 18.722000122070312, ...],
203
+ 'segment_speakers': ['A', 'B', ...],
204
+ 'words', ["hmm", "hmm", ...]
205
+ 'channels': [0, 0, ..],
206
+ 'file': "/.cache/huggingface/datasets/downloads/af7e748544004557b35eef8b0522d4fb2c71e004b82ba8b7343913a15def465f"
207
+ }
208
+ ```
209
+
210
+ ### Data Fields
211
+
212
+ - word_ids: a list of the ids of the words
213
+
214
+ - word_start_times: a list of the start times of when the words were spoken in seconds
215
+
216
+ - word_end_times: a list of the end times of when the words were spoken in seconds
217
+
218
+ - word_speakers: a list of speakers one for each word
219
+
220
+ - segment_ids: a list of the ids of the segments
221
+
222
+ - segment_start_times: a list of the start times of when the segments start
223
+
224
+ - segment_end_times: a list of the start times of when the segments ends
225
+
226
+ - segment_speakers: a list of speakers one for each segment
227
+
228
+ - words: a list of all the spoken words
229
+
230
+ - channels: a list of all channels that were used for each word
231
+
232
+ - file: a path to the audio file
233
+
234
+ ### Data Splits
235
+
236
+ The dataset consists of several configurations, each one having train/validation/test splits:
237
+
238
+ - headset-single: Close talking audio of single headset. This configuration only includes audio belonging to the headset of the person currently speaking.
239
+
240
+ - headset-multi (4 channels): Close talking audio of four individual headset. This configuration includes audio belonging to four individual headsets. For each annotation there are 4 audio files 0, 1, 2, 3.
241
+
242
+ - microphone-single: Far field audio of single microphone. This configuration only includes audio belonging the first microphone, *i.e.* 1-1, of the microphone array.
243
+
244
+ - microphone-multi (8 channels): Far field audio of microphone array. This configuration includes audio of the first microphone array 1-1, 1-2, ..., 1-8.
245
+
246
+ In general, `headset-single` and `headset-multi` include significantly less noise than
247
+ `microphone-single` and `microphone-multi`.
248
+
249
+ | | Train | Valid | Test |
250
+ | ----- | ------ | ----- | ---- |
251
+ | headset-single | 136 (80h) | 18 (9h) | 16 (9h) |
252
+ | headset-multi (4 channels) | 136 (320h) | 18 (36h) | 16 (36h) |
253
+ | microphone-single | 136 (80h) | 18 (9h) | 16 (9h) |
254
+ | microphone-multi (8 channels) | 136 (640h) | 18 (72h) | 16 (72h) |
255
+
256
+ Note that each sample contains between 10 and 60 minutes of audio data which makes it
257
+ impractical for direct transcription. One should make use of the segment and word start times and end times to chunk the samples into smaller samples of manageable size.
258
+
259
+ ## Dataset Creation
260
+
261
+ All information about the dataset creation can be found
262
+ [here](https://groups.inf.ed.ac.uk/ami/corpus/overview.shtml)
263
+
264
+ ### Curation Rationale
265
+
266
+ [Needs More Information]
267
+
268
+ ### Source Data
269
+
270
+ #### Initial Data Collection and Normalization
271
+
272
+ [Needs More Information]
273
+
274
+ #### Who are the source language producers?
275
+
276
+ [Needs More Information]
277
+
278
+ ### Annotations
279
+
280
+ #### Annotation process
281
+
282
+ [Needs More Information]
283
+
284
+ #### Who are the annotators?
285
+
286
+ [Needs More Information]
287
+
288
+ ### Personal and Sensitive Information
289
+
290
+ [Needs More Information]
291
+
292
+ ## Considerations for Using the Data
293
+
294
+ ### Social Impact of Dataset
295
+
296
+ [More Information Needed]
297
+
298
+ ### Discussion of Biases
299
+
300
+ [More Information Needed]
301
+
302
+ ### Other Known Limitations
303
+
304
+ [Needs More Information]
305
+
306
+ ## Additional Information
307
+
308
+ ### Dataset Curators
309
+
310
+ [Needs More Information]
311
+
312
+ ### Licensing Information
313
+
314
+ CC BY 4.0
315
+
316
+ ### Citation Information
317
+ #### TODO
318
+
319
+ ### Contributions
320
+
321
+ Thanks to [@cahya-wirawan](https://github.com/cahya-wirawan) and [@patrickvonplaten](https://github.com/patrickvonplaten) for adding this dataset.
322
+ #### TODO
ami.py ADDED
@@ -0,0 +1,581 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2021 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """AMI Corpus"""
18
+
19
+ import os
20
+ import xml.etree.ElementTree as ET
21
+
22
+ import numpy as np
23
+
24
+ import datasets
25
+
26
+
27
+ logger = datasets.logging.get_logger(__name__)
28
+
29
+ _CITATION = """\
30
+ @inproceedings{10.1007/11677482_3,
31
+ author = {Carletta, Jean and Ashby, Simone and Bourban, Sebastien and Flynn, Mike and Guillemot, Mael and Hain, Thomas and Kadlec, Jaroslav and Karaiskos, Vasilis and Kraaij, Wessel and Kronenthal, Melissa and Lathoud, Guillaume and Lincoln, Mike and Lisowska, Agnes and McCowan, Iain and Post, Wilfried and Reidsma, Dennis and Wellner, Pierre},
32
+ title = {The AMI Meeting Corpus: A Pre-Announcement},
33
+ year = {2005},
34
+ isbn = {3540325492},
35
+ publisher = {Springer-Verlag},
36
+ address = {Berlin, Heidelberg},
37
+ url = {https://doi.org/10.1007/11677482_3},
38
+ doi = {10.1007/11677482_3},
39
+ abstract = {The AMI Meeting Corpus is a multi-modal data set consisting of 100 hours of meeting
40
+ recordings. It is being created in the context of a project that is developing meeting
41
+ browsing technology and will eventually be released publicly. Some of the meetings
42
+ it contains are naturally occurring, and some are elicited, particularly using a scenario
43
+ in which the participants play different roles in a design team, taking a design project
44
+ from kick-off to completion over the course of a day. The corpus is being recorded
45
+ using a wide range of devices including close-talking and far-field microphones, individual
46
+ and room-view video cameras, projection, a whiteboard, and individual pens, all of
47
+ which produce output signals that are synchronized with each other. It is also being
48
+ hand-annotated for many different phenomena, including orthographic transcription,
49
+ discourse properties such as named entities and dialogue acts, summaries, emotions,
50
+ and some head and hand gestures. We describe the data set, including the rationale
51
+ behind using elicited material, and explain how the material is being recorded, transcribed
52
+ and annotated.},
53
+ booktitle = {Proceedings of the Second International Conference on Machine Learning for Multimodal Interaction},
54
+ pages = {28–39},
55
+ numpages = {12},
56
+ location = {Edinburgh, UK},
57
+ series = {MLMI'05}
58
+ }
59
+ """
60
+
61
+ _URL = "https://groups.inf.ed.ac.uk/ami/corpus/"
62
+
63
+ _DL_URL_ANNOTATIONS = "http://groups.inf.ed.ac.uk/ami/AMICorpusAnnotations/ami_public_manual_1.6.2.zip"
64
+ _DL_SAMPLE_FORMAT = "https://groups.inf.ed.ac.uk/ami/AMICorpusMirror//amicorpus/{}/audio/{}"
65
+
66
+ _SPEAKERS = ["A", "B", "C", "D", "E"]
67
+
68
+ # Commented out samples don't seem to exist
69
+
70
+ _TRAIN_SAMPLE_IDS = [
71
+ "ES2002a",
72
+ "ES2002b",
73
+ "ES2002c",
74
+ "ES2002d",
75
+ "ES2003a",
76
+ "ES2003b",
77
+ "ES2003c",
78
+ "ES2003d",
79
+ "ES2005a",
80
+ "ES2005b",
81
+ "ES2005c",
82
+ "ES2005d",
83
+ "ES2006a",
84
+ "ES2006b",
85
+ "ES2006c",
86
+ "ES2006d",
87
+ "ES2007a",
88
+ "ES2007b",
89
+ "ES2007c",
90
+ "ES2007d",
91
+ "ES2008a",
92
+ "ES2008b",
93
+ "ES2008c",
94
+ "ES2008d",
95
+ "ES2009a",
96
+ "ES2009b",
97
+ "ES2009c",
98
+ "ES2009d",
99
+ "ES2010a",
100
+ "ES2010b",
101
+ "ES2010c",
102
+ "ES2010d",
103
+ "ES2012a",
104
+ "ES2012b",
105
+ "ES2012c",
106
+ "ES2012d",
107
+ "ES2013a",
108
+ "ES2013b",
109
+ "ES2013c",
110
+ "ES2013d",
111
+ "ES2014a",
112
+ "ES2014b",
113
+ "ES2014c",
114
+ "ES2014d",
115
+ "ES2015a",
116
+ "ES2015b",
117
+ "ES2015c",
118
+ "ES2015d",
119
+ "ES2016a",
120
+ "ES2016b",
121
+ "ES2016c",
122
+ "ES2016d",
123
+ "IS1000a",
124
+ "IS1000b",
125
+ "IS1000c",
126
+ "IS1000d",
127
+ "IS1001a",
128
+ "IS1001b",
129
+ "IS1001c",
130
+ "IS1001d",
131
+ "IS1002b",
132
+ "IS1002c",
133
+ "IS1002d",
134
+ "IS1003a",
135
+ "IS1003b",
136
+ "IS1003c",
137
+ "IS1003d",
138
+ "IS1004a",
139
+ "IS1004b",
140
+ "IS1004c",
141
+ "IS1004d",
142
+ "IS1005a",
143
+ "IS1005b",
144
+ "IS1005c",
145
+ "IS1006a",
146
+ "IS1006b",
147
+ "IS1006c",
148
+ "IS1006d",
149
+ "IS1007a",
150
+ "IS1007b",
151
+ "IS1007c",
152
+ "IS1007d",
153
+ "TS3005a",
154
+ "TS3005b",
155
+ "TS3005c",
156
+ "TS3005d",
157
+ "TS3006a",
158
+ "TS3006b",
159
+ "TS3006c",
160
+ "TS3006d",
161
+ "TS3007a",
162
+ "TS3007b",
163
+ "TS3007c",
164
+ "TS3007d",
165
+ "TS3008a",
166
+ "TS3008b",
167
+ "TS3008c",
168
+ "TS3008d",
169
+ "TS3009a",
170
+ "TS3009b",
171
+ "TS3009c",
172
+ "TS3009d",
173
+ "TS3010a",
174
+ "TS3010b",
175
+ "TS3010c",
176
+ "TS3010d",
177
+ "TS3011a",
178
+ "TS3011b",
179
+ "TS3011c",
180
+ "TS3011d",
181
+ "TS3012a",
182
+ "TS3012b",
183
+ "TS3012c",
184
+ "TS3012d",
185
+ "EN2001a",
186
+ "EN2001b",
187
+ "EN2001d",
188
+ "EN2001e",
189
+ "EN2003a",
190
+ "EN2004a",
191
+ "EN2005a",
192
+ "EN2006a",
193
+ "EN2006b",
194
+ "EN2009b",
195
+ "EN2009c",
196
+ "EN2009d",
197
+ "IN1001",
198
+ "IN1002",
199
+ "IN1005",
200
+ "IN1007",
201
+ "IN1008",
202
+ "IN1009",
203
+ "IN1012",
204
+ "IN1013",
205
+ "IN1014",
206
+ "IN1016",
207
+ ]
208
+
209
+ _VALIDATION_SAMPLE_IDS = [
210
+ "ES2011a",
211
+ "ES2011b",
212
+ "ES2011c",
213
+ "ES2011d",
214
+ "IS1008a",
215
+ "IS1008b",
216
+ "IS1008c",
217
+ "IS1008d",
218
+ "TS3004a",
219
+ "TS3004b",
220
+ "TS3004c",
221
+ "TS3004d",
222
+ "IB4001",
223
+ "IB4002",
224
+ "IB4003",
225
+ "IB4004",
226
+ "IB4010",
227
+ "IB4011",
228
+ ]
229
+
230
+
231
+ _EVAL_SAMPLE_IDS = [
232
+ "ES2004a",
233
+ "ES2004b",
234
+ "ES2004c",
235
+ "ES2004d",
236
+ "IS1009a",
237
+ "IS1009b",
238
+ "IS1009c",
239
+ "IS1009d",
240
+ "TS3003a",
241
+ "TS3003b",
242
+ "TS3003c",
243
+ "TS3003d",
244
+ "EN2002a",
245
+ "EN2002b",
246
+ "EN2002c",
247
+ "EN2002d",
248
+ ]
249
+
250
+
251
+ _DESCRIPTION = """\
252
+ The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
253
+ synchronized to a common timeline. These include close-talking and far-field microphones, individual and
254
+ room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
255
+ the participants also have unsynchronized pens available to them that record what is written. The meetings
256
+ were recorded in English using three different rooms with different acoustic properties, and include mostly
257
+ non-native speakers. \n
258
+ """
259
+
260
+
261
+ class AMIConfig(datasets.BuilderConfig):
262
+ """BuilderConfig for LibriSpeechASR."""
263
+
264
+ def __init__(self, formats, missing_files=None, **kwargs):
265
+ """
266
+ Args:
267
+ formats: `List[string]`, a list of audio file formats
268
+ missing_files: `List[string]`, a list of missing audio file ids
269
+ **kwargs: keyword arguments forwarded to super.
270
+ """
271
+ self.dl_path_formats = [_DL_SAMPLE_FORMAT + "." + f + ".wav" for f in formats]
272
+
273
+ # for microphone configs some audio files are missing
274
+ self.missing_files = missing_files if missing_files is not None else []
275
+ super(AMIConfig, self).__init__(version=datasets.Version("1.6.2", ""), **kwargs)
276
+
277
+
278
+ class AMI(datasets.GeneratorBasedBuilder):
279
+ """AMI dataset."""
280
+
281
+ BUILDER_CONFIGS = [
282
+ AMIConfig(name="headset-single", formats=["Mix-Headset"], description=""),
283
+ AMIConfig(name="headset-multi", formats=["Headset-0", "Headset-1", "Headset-2", "Headset-3"], description=""),
284
+ AMIConfig(
285
+ name="microphone-single",
286
+ formats=["Array1-01"],
287
+ missing_files=["IS1003b", "IS1007d"],
288
+ ),
289
+ AMIConfig(
290
+ name="microphone-multi",
291
+ formats=[
292
+ "Array1-01",
293
+ "Array1-02",
294
+ "Array1-03",
295
+ "Array1-04",
296
+ "Array1-05",
297
+ "Array1-06",
298
+ "Array1-07",
299
+ "Array1-08",
300
+ ],
301
+ missing_files=["IS1003b", "IS1007d"],
302
+ ),
303
+ ]
304
+
305
+ def _info(self):
306
+ features_dict = {
307
+ "word_ids": datasets.Sequence(datasets.Value("string")),
308
+ "word_start_times": datasets.Sequence(datasets.Value("float")),
309
+ "word_end_times": datasets.Sequence(datasets.Value("float")),
310
+ "word_speakers": datasets.Sequence(datasets.Value("string")),
311
+ "segment_ids": datasets.Sequence(datasets.Value("string")),
312
+ "segment_start_times": datasets.Sequence(datasets.Value("float")),
313
+ "segment_end_times": datasets.Sequence(datasets.Value("float")),
314
+ "segment_speakers": datasets.Sequence(datasets.Value("string")),
315
+ "words": datasets.Sequence(datasets.Value("string")),
316
+ "channels": datasets.Sequence(datasets.Value("string")),
317
+ }
318
+
319
+ if self.config.name == "headset-single":
320
+ features_dict.update({"file": datasets.Value("string")})
321
+ config_description = (
322
+ "Close talking audio of single headset. "
323
+ "This configuration only includes audio belonging to the "
324
+ "headset of the person currently speaking."
325
+ )
326
+ elif self.config.name == "microphone-single":
327
+ features_dict.update({"file": datasets.Value("string")})
328
+ config_description = (
329
+ "Far field audio of single microphone. "
330
+ "This configuration only includes audio belonging the first microphone, "
331
+ "*i.e.* 1-1, of the microphone array."
332
+ )
333
+ elif self.config.name == "headset-multi":
334
+ features_dict.update(
335
+ {
336
+ "file-0": datasets.Value("string"),
337
+ "file-1": datasets.Value("string"),
338
+ "file-2": datasets.Value("string"),
339
+ "file-3": datasets.Value("string"),
340
+ }
341
+ )
342
+ config_description = (
343
+ "Close talking audio of four individual headset. "
344
+ "This configuration includes audio belonging to four individual headsets."
345
+ " For each annotation there are 4 audio files 0, 1, 2, 3."
346
+ )
347
+ elif self.config.name == "microphone-multi":
348
+ features_dict.update(
349
+ {
350
+ "file-1-1": datasets.Value("string"),
351
+ "file-1-2": datasets.Value("string"),
352
+ "file-1-3": datasets.Value("string"),
353
+ "file-1-4": datasets.Value("string"),
354
+ "file-1-5": datasets.Value("string"),
355
+ "file-1-6": datasets.Value("string"),
356
+ "file-1-7": datasets.Value("string"),
357
+ "file-1-8": datasets.Value("string"),
358
+ }
359
+ )
360
+ config_description = (
361
+ "Far field audio of microphone array. "
362
+ "This configuration includes audio of "
363
+ "the first microphone array 1-1, 1-2, ..., 1-8."
364
+ )
365
+ else:
366
+ raise ValueError(f"Configuration {self.config.name} does not exist.")
367
+
368
+ return datasets.DatasetInfo(
369
+ description=_DESCRIPTION + config_description,
370
+ features=datasets.Features(features_dict),
371
+ homepage=_URL,
372
+ citation=_CITATION,
373
+ )
374
+
375
+ def _split_generators(self, dl_manager):
376
+
377
+ # multi-processing doesn't work for AMI
378
+ if hasattr(dl_manager, "_download_config") and dl_manager._download_config.num_proc != 1:
379
+ logger.warning(
380
+ "AMI corpus cannot be downloaded using multi-processing. "
381
+ "Setting number of downloaded processes `num_proc` to 1. "
382
+ )
383
+ dl_manager._download_config.num_proc = 1
384
+
385
+ annotation_path = dl_manager.download_and_extract(_DL_URL_ANNOTATIONS)
386
+
387
+ # train
388
+ train_files = [path.format(_id, _id) for _id in _TRAIN_SAMPLE_IDS for path in self.config.dl_path_formats]
389
+ train_files = list(
390
+ filter(lambda f: f.split("/")[-1].split(".")[0] not in self.config.missing_files, train_files)
391
+ )
392
+ train_ids = [f.split("/")[-1].split(".")[0] for f in train_files]
393
+ train_path = dl_manager.download_and_extract(train_files)
394
+
395
+ # validation
396
+ validation_files = [
397
+ path.format(_id, _id) for _id in _VALIDATION_SAMPLE_IDS for path in self.config.dl_path_formats
398
+ ]
399
+ validation_ids = [f.split("/")[-1].split(".")[0] for f in validation_files]
400
+ validation_path = dl_manager.download_and_extract(validation_files)
401
+
402
+ # test
403
+ eval_files = [path.format(_id, _id) for _id in _EVAL_SAMPLE_IDS for path in self.config.dl_path_formats]
404
+ eval_ids = [f.split("/")[-1].split(".")[0] for f in eval_files]
405
+ eval_path = dl_manager.download_and_extract(eval_files)
406
+
407
+ return [
408
+ datasets.SplitGenerator(
409
+ name=datasets.Split.TRAIN,
410
+ gen_kwargs={
411
+ "annotation_path": annotation_path,
412
+ "samples_paths": train_path,
413
+ "ids": _TRAIN_SAMPLE_IDS,
414
+ "verification_ids": train_ids,
415
+ },
416
+ ),
417
+ datasets.SplitGenerator(
418
+ name=datasets.Split.VALIDATION,
419
+ gen_kwargs={
420
+ "annotation_path": annotation_path,
421
+ "samples_paths": validation_path,
422
+ "ids": _VALIDATION_SAMPLE_IDS,
423
+ "verification_ids": validation_ids,
424
+ },
425
+ ),
426
+ datasets.SplitGenerator(
427
+ name=datasets.Split.TEST,
428
+ gen_kwargs={
429
+ "annotation_path": annotation_path,
430
+ "samples_paths": eval_path,
431
+ "ids": _EVAL_SAMPLE_IDS,
432
+ "verification_ids": eval_ids,
433
+ },
434
+ ),
435
+ ]
436
+
437
+ @staticmethod
438
+ def _sort(key, lists):
439
+ indices = np.argsort(key)
440
+
441
+ sorted_lists = [np.array(array)[indices].tolist() for array in lists]
442
+ return sorted_lists
443
+
444
+ @staticmethod
445
+ def _extract_words_annotations(paths):
446
+ word_ids = []
447
+ word_start_times = []
448
+ word_end_times = []
449
+ words = []
450
+ word_speakers = []
451
+
452
+ for path in paths:
453
+ # retrive speaker
454
+ speaker = path.split(".")[-3]
455
+
456
+ with open(path, "r", encoding="utf-8") as words_file:
457
+ root = ET.parse(words_file).getroot()
458
+ for type_tag in root.findall("w"):
459
+ word_id = type_tag.get("{http://nite.sourceforge.net/}id")
460
+
461
+ word_start_time = type_tag.get("starttime")
462
+ word_end_time = type_tag.get("endtime")
463
+ text = type_tag.text
464
+
465
+ if word_start_time is not None and word_end_time is not None:
466
+ word_ids.append(word_id)
467
+ word_start_times.append(float(word_start_time))
468
+ word_end_times.append(float(word_end_time))
469
+ words.append(text)
470
+ word_speakers.append(speaker)
471
+ else:
472
+ logger.warning(
473
+ f"Annotation {word_id} of file {path} is missing information about"
474
+ "either word_start_time or word_end_time. Skipping sample..."
475
+ )
476
+
477
+ return AMI._sort(word_start_times, [word_ids, word_start_times, word_end_times, words, word_speakers])
478
+
479
+ @staticmethod
480
+ def _extract_segments_annotations(paths):
481
+ segment_ids = []
482
+ channels = []
483
+ segment_start_times = []
484
+ segment_end_times = []
485
+ segment_speakers = []
486
+
487
+ for path in paths:
488
+ speaker = path.split(".")[-3]
489
+
490
+ with open(path, "r", encoding="utf-8") as segments_file:
491
+ root = ET.parse(segments_file).getroot()
492
+ for type_tag in root.findall("segment"):
493
+ segment_ids.append(type_tag.get("{http://nite.sourceforge.net/}id"))
494
+ segment_start_times.append(float(type_tag.get("transcriber_start")))
495
+ segment_end_times.append(float(type_tag.get("transcriber_end")))
496
+ channels.append(type_tag.get("channel"))
497
+ segment_speakers.append(speaker)
498
+
499
+ return AMI._sort(
500
+ segment_start_times, [segment_ids, segment_start_times, segment_end_times, channels, segment_speakers]
501
+ )
502
+
503
+ def _generate_examples(self, annotation_path, samples_paths, ids, verification_ids):
504
+ logger.info(f"⏳ Generating {', '.join(ids)}")
505
+
506
+ # number of audio files of config
507
+ num_audios = len(self.config.dl_path_formats)
508
+
509
+ # filter missing ids
510
+ ids = list(filter(lambda n: n not in self.config.missing_files, ids))
511
+
512
+ # audio
513
+ samples_paths_dict = {}
514
+ for i, _id in enumerate(ids):
515
+ sample_paths = samples_paths[num_audios * i : num_audios * (i + 1)]
516
+ sample_verification_id = set(verification_ids[num_audios * i : num_audios * (i + 1)])
517
+
518
+ # make sure that multi microphone samples are correctly atttributed to labels
519
+ if len(sample_verification_id) > 1 or next(iter(sample_verification_id)) != _id:
520
+ raise ValueError(
521
+ f"Incorrect dataset generation. The files {sample_paths} of id {_id} have incorrect verification_ids {sample_verification_id}."
522
+ )
523
+
524
+ # set correct files correctly
525
+ samples_paths_dict[_id] = sample_paths
526
+
527
+ # words
528
+ words_paths = {
529
+ _id: [os.path.join(annotation_path, "words/{}.{}.words.xml".format(_id, speaker)) for speaker in _SPEAKERS]
530
+ for _id in ids
531
+ }
532
+ words_paths = {_id: list(filter(lambda path: os.path.isfile(path), words_paths[_id])) for _id in ids}
533
+ words_paths = {key: words_paths[key] for key in words_paths if len(words_paths[key]) > 0}
534
+
535
+ # segments
536
+ segments_paths = {
537
+ _id: [
538
+ os.path.join(annotation_path, "segments/{}.{}.segments.xml".format(_id, speaker))
539
+ for speaker in _SPEAKERS
540
+ ]
541
+ for _id in ids
542
+ }
543
+ segments_paths = {_id: list(filter(lambda path: os.path.isfile(path), segments_paths[_id])) for _id in ids}
544
+ segments_paths = {key: segments_paths[key] for key in segments_paths if len(segments_paths[key]) > 0}
545
+
546
+ for _id in words_paths.keys():
547
+ word_ids, word_start_times, word_end_times, words, word_speakers = self._extract_words_annotations(
548
+ words_paths[_id]
549
+ )
550
+
551
+ (
552
+ segment_ids,
553
+ segment_start_times,
554
+ segment_end_times,
555
+ channels,
556
+ segment_speakers,
557
+ ) = self._extract_segments_annotations(segments_paths[_id])
558
+
559
+ result = {
560
+ "word_ids": word_ids,
561
+ "word_start_times": word_start_times,
562
+ "word_end_times": word_end_times,
563
+ "word_speakers": word_speakers,
564
+ "segment_ids": segment_ids,
565
+ "segment_start_times": segment_start_times,
566
+ "segment_end_times": segment_end_times,
567
+ "segment_speakers": segment_speakers,
568
+ "channels": channels,
569
+ "words": words,
570
+ }
571
+
572
+ if self.config.name in ["headset-single", "microphone-single"]:
573
+ result.update({"file": samples_paths_dict[_id][0]})
574
+ elif self.config.name in ["headset-multi"]:
575
+ result.update({f"file-{i}": samples_paths_dict[_id][i] for i in range(num_audios)})
576
+ elif self.config.name in ["microphone-multi"]:
577
+ result.update({f"file-1-{i+1}": samples_paths_dict[_id][i] for i in range(num_audios)})
578
+ else:
579
+ raise ValueError(f"Configuration {self.config.name} does not exist.")
580
+
581
+ yield _id, result
dataset_infos.json ADDED
The diff for this file is too large to render. See raw diff
 
dummy/headset-multi/1.6.2/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ec3d9c11ef6c29bdf89592e5ae3b88b2336ff055ab362be138c5827c5999dee
3
+ size 48550
dummy/headset-single/1.6.2/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ec3d9c11ef6c29bdf89592e5ae3b88b2336ff055ab362be138c5827c5999dee
3
+ size 48550
dummy/microphone-multi/1.6.2/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ec3d9c11ef6c29bdf89592e5ae3b88b2336ff055ab362be138c5827c5999dee
3
+ size 48550
dummy/microphone-single/1.6.2/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ec3d9c11ef6c29bdf89592e5ae3b88b2336ff055ab362be138c5827c5999dee
3
+ size 48550