polinaeterna HF staff commited on
Commit
46e652e
1 Parent(s): f8f30d5

add more info to readme

Browse files
Files changed (1) hide show
  1. README.md +116 -11
README.md CHANGED
@@ -1,13 +1,56 @@
1
- # AMI Corpus
2
-
3
- https://groups.inf.ed.ac.uk/ami/corpus/
4
-
5
- To be filled!
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
 
7
  **Note**: This dataset corresponds to the data-processing of [KALDI's AMI S5 recipe](https://github.com/kaldi-asr/kaldi/tree/master/egs/ami/s5).
8
  This means text is normalized and the audio data is chunked according to the scripts above!
9
  To make the user experience as simply as possible, we provide the already chunked data to the user here so that the following can be done:
10
 
 
 
 
11
  ```python
12
  from datasets import load_dataset
13
  ds = load_dataset("edinburghcstr/ami", "ihm")
@@ -18,15 +61,15 @@ gives:
18
  ```
19
  DatasetDict({
20
  train: Dataset({
21
- features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
22
  num_rows: 108502
23
  })
24
  validation: Dataset({
25
- features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
26
  num_rows: 13098
27
  })
28
  test: Dataset({
29
- features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
30
  num_rows: 12643
31
  })
32
  })
@@ -39,10 +82,10 @@ ds["train"][0]
39
  automatically loads the audio into memory:
40
 
41
  ```
42
- {'segment_id': 'EN2001a',
43
  'audio_id': 'AMI_EN2001a_H00_MEE068_0000557_0000594',
44
  'text': 'OKAY',
45
- 'audio': {'path': '/home/patrick_huggingface_co/.cache/huggingface/datasets/downloads/extracted/2d75d5b3e8a91f44692e2973f08b4cac53698f92c2567bd43b41d19c313a5280/EN2001a/train_ami_en2001a_h00_mee068_0000557_0000594.wav',
46
  'array': array([0. , 0. , 0. , ..., 0.00033569, 0.00030518,
47
  0.00030518], dtype=float32),
48
  'sampling_rate': 16000},
@@ -70,4 +113,66 @@ The results are in-line with results of published papers:
70
  - [*Hybrid acoustic models for distant and multichannel large vocabulary speech recognition*](https://www.researchgate.net/publication/258075865_Hybrid_acoustic_models_for_distant_and_multichannel_large_vocabulary_speech_recognition)
71
  - [Multi-Span Acoustic Modelling using Raw Waveform Signals](https://arxiv.org/abs/1906.11047)
72
 
73
- You can run [run.sh](https://huggingface.co/patrickvonplaten/ami-wav2vec2-large-lv60/blob/main/run.sh) to reproduce the result.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Dataset Card for AMI
2
+
3
+ ## Table of Contents
4
+ - [Table of Contents](#table-of-contents)
5
+ - [Dataset Description](#dataset-description)
6
+ - [Dataset Summary](#dataset-summary)
7
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
8
+ - [Languages](#languages)
9
+ - [Dataset Structure](#dataset-structure)
10
+ - [Data Instances](#data-instances)
11
+ - [Data Fields](#data-fields)
12
+ - [Data Splits](#data-splits)
13
+ - [Dataset Creation](#dataset-creation)
14
+ - [Curation Rationale](#curation-rationale)
15
+ - [Source Data](#source-data)
16
+ - [Annotations](#annotations)
17
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
18
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
19
+ - [Social Impact of Dataset](#social-impact-of-dataset)
20
+ - [Discussion of Biases](#discussion-of-biases)
21
+ - [Other Known Limitations](#other-known-limitations)
22
+ - [Additional Information](#additional-information)
23
+ - [Dataset Curators](#dataset-curators)
24
+ - [Licensing Information](#licensing-information)
25
+ - [Citation Information](#citation-information)
26
+ - [Contributions](#contributions)
27
+ - [Terms of Usage](#terms-of-usage)
28
+
29
+
30
+ ## Dataset Description
31
+
32
+ - **Homepage:** https://groups.inf.ed.ac.uk/ami/corpus/
33
+ - **Repository:** https://github.com/kaldi-asr/kaldi/tree/master/egs/ami/s5
34
+ - **Paper:**
35
+ - **Leaderboard:**
36
+ - **Point of Contact:** [jonathan@ed.ac.uk](mailto:jonathan@ed.ac.uk)
37
+
38
+ ## Dataset Description
39
+
40
+ The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
41
+ synchronized to a common timeline. These include close-talking and far-field microphones, individual and
42
+ room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
43
+ the participants also have unsynchronized pens available to them that record what is written. The meetings
44
+ were recorded in English using three different rooms with different acoustic properties, and include mostly
45
+ non-native speakers.
46
 
47
  **Note**: This dataset corresponds to the data-processing of [KALDI's AMI S5 recipe](https://github.com/kaldi-asr/kaldi/tree/master/egs/ami/s5).
48
  This means text is normalized and the audio data is chunked according to the scripts above!
49
  To make the user experience as simply as possible, we provide the already chunked data to the user here so that the following can be done:
50
 
51
+
52
+ ### Example Usage
53
+
54
  ```python
55
  from datasets import load_dataset
56
  ds = load_dataset("edinburghcstr/ami", "ihm")
 
61
  ```
62
  DatasetDict({
63
  train: Dataset({
64
+ features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
65
  num_rows: 108502
66
  })
67
  validation: Dataset({
68
+ features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
69
  num_rows: 13098
70
  })
71
  test: Dataset({
72
+ features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
73
  num_rows: 12643
74
  })
75
  })
 
82
  automatically loads the audio into memory:
83
 
84
  ```
85
+ {'meeting_id': 'EN2001a',
86
  'audio_id': 'AMI_EN2001a_H00_MEE068_0000557_0000594',
87
  'text': 'OKAY',
88
+ 'audio': {'path': '/cache/dir/path/downloads/extracted/2d75d5b3e8a91f44692e2973f08b4cac53698f92c2567bd43b41d19c313a5280/EN2001a/train_ami_en2001a_h00_mee068_0000557_0000594.wav',
89
  'array': array([0. , 0. , 0. , ..., 0.00033569, 0.00030518,
90
  0.00030518], dtype=float32),
91
  'sampling_rate': 16000},
 
113
  - [*Hybrid acoustic models for distant and multichannel large vocabulary speech recognition*](https://www.researchgate.net/publication/258075865_Hybrid_acoustic_models_for_distant_and_multichannel_large_vocabulary_speech_recognition)
114
  - [Multi-Span Acoustic Modelling using Raw Waveform Signals](https://arxiv.org/abs/1906.11047)
115
 
116
+ You can run [run.sh](https://huggingface.co/patrickvonplaten/ami-wav2vec2-large-lv60/blob/main/run.sh) to reproduce the result.
117
+
118
+ ### Supported Tasks and Leaderboards
119
+
120
+ ### Languages
121
+
122
+ ## Dataset Structure
123
+
124
+ ### Data Instances
125
+
126
+ ### Data Fields
127
+
128
+ ### Data Splits
129
+
130
+ #### Transcribed Subsets Size
131
+
132
+ ## Dataset Creation
133
+
134
+ ### Curation Rationale
135
+
136
+ ### Source Data
137
+
138
+ #### Initial Data Collection and Normalization
139
+
140
+ #### Who are the source language producers?
141
+
142
+ ### Annotations
143
+
144
+ #### Annotation process
145
+
146
+ #### Who are the annotators?
147
+
148
+ ### Personal and Sensitive Information
149
+
150
+ ## Considerations for Using the Data
151
+
152
+ ### Social Impact of Dataset
153
+
154
+ [More Information Needed]
155
+
156
+ ### Discussion of Biases
157
+
158
+ ### Other Known Limitations
159
+
160
+ ## Additional Information
161
+
162
+ ### Dataset Curators
163
+
164
+
165
+ ### Licensing Information
166
+
167
+
168
+ ### Citation Information
169
+
170
+
171
+ ### Contributions
172
+
173
+ Thanks to [@sanchit-gandhi](https://github.com/sanchit-gandhi), [@patrickvonplaten](https://github.com/patrickvonplaten),
174
+ and [@polinaeterna](https://github.com/polinaeterna) for adding this dataset.
175
+
176
+ ## Terms of Usage
177
+
178
+