m-wosik commited on
Commit
3165c67
1 Parent(s): d81d389

Update README.md with basic, high-level information

Browse files
Files changed (1) hide show
  1. README.md +65 -19
README.md CHANGED
@@ -2,27 +2,30 @@
2
  annotations_creators:
3
  - expert-generated
4
  language:
5
- - pl
 
 
 
 
6
  license:
7
  - mit
8
  multilinguality:
9
  - monolingual
10
-
11
  dataset_info:
12
- - config_name: config
13
- features:
14
- - name: audio_id
15
- dtype: string
16
- - name: audio
17
- dtype:
18
- audio:
19
- sampling_rate: 16000
20
- - name: text
21
- dtype: string
22
  ---
23
 
24
 
25
- # Dataset Card for [Dataset Name]
26
 
27
  ## Table of Contents
28
  - [Table of Contents](#table-of-contents)
@@ -59,15 +62,36 @@ dataset_info:
59
 
60
  ### Dataset Summary
61
 
62
- [More Information Needed]
 
 
 
 
 
 
 
 
 
 
63
 
64
  ### Supported Tasks and Leaderboards
65
 
66
- [More Information Needed]
 
 
 
 
 
 
67
 
68
  ### Languages
69
 
70
- [More Information Needed]
 
 
 
 
 
71
 
72
  ## Dataset Structure
73
 
@@ -81,10 +105,25 @@ dataset_info:
81
 
82
  ### Data Splits
83
 
84
- [More Information Needed]
 
 
 
 
 
85
 
86
  ## Dataset Creation
87
 
 
 
 
 
 
 
 
 
 
 
88
  ### Curation Rationale
89
 
90
  [More Information Needed]
@@ -121,7 +160,7 @@ dataset_info:
121
 
122
  ### Discussion of Biases
123
 
124
- [More Information Needed]
125
 
126
  ### Other Known Limitations
127
 
@@ -139,7 +178,14 @@ dataset_info:
139
 
140
  ### Citation Information
141
 
142
- [More Information Needed]
 
 
 
 
 
 
 
143
 
144
  ### Contributions
145
 
 
2
  annotations_creators:
3
  - expert-generated
4
  language:
5
+ - en
6
+ - de
7
+ - es
8
+ - fr
9
+ - it
10
  license:
11
  - mit
12
  multilinguality:
13
  - monolingual
 
14
  dataset_info:
15
+ - config_name: config
16
+ features:
17
+ - name: audio_id
18
+ dtype: string
19
+ - name: audio
20
+ dtype:
21
+ audio:
22
+ sampling_rate: 16000
23
+ - name: text
24
+ dtype: string
25
  ---
26
 
27
 
28
+ # MOCKS dataset
29
 
30
  ## Table of Contents
31
  - [Table of Contents](#table-of-contents)
 
62
 
63
  ### Dataset Summary
64
 
65
+ Multilingual Open Custom Keyword Spotting Testset (MOCKS) is a comprehensive audio testset for evaluation and benchmarking
66
+ Open-Vocabulary Keyword Spotting (OV-KWS) models. It supports multiple OV-KWS problems:
67
+ both text-based and audio-based keyword spotting, as well as offline and online (streaming) modes.
68
+ It is based on the LibriSpeech and Mozilla Common Voice datasets and contains
69
+ almost 50,000 keywords, with audio data available in English, French, German, Italian, and Spanish.
70
+ The testset was generated using automatically generated alignments used for the extraction of parts of the recordings that were split into keywords and test samples.
71
+ MOCKS contains both positive and negative examples selected based on phonetic transcriptions that are challenging and should allow for in-depth OV-KWS model evaluation.
72
+
73
+ Please refer to our [paper]() for further details.
74
+
75
+ [More Information Needed - add link to paper]
76
 
77
  ### Supported Tasks and Leaderboards
78
 
79
+ The MOCKS dataset can be used for Open-Vocabulary Keyword Spotting (OV-KWS) task. It supports two OV-KWS types:
80
+ - Query-by-Text, where keyword is provided by text and needs to be detected on audio stream.
81
+ - Query-by-Example, where keyword is provided with enrollment audio for detection on audio stream.
82
+
83
+ It also allows for:
84
+ - offline keyword detection, where test audio is trimed to contrain only keyword of interest.
85
+ - online (streaming) keyword detection, where test audio have past and future context besides keyword of interest.
86
 
87
  ### Languages
88
 
89
+ The MOCKS incorporates 5 languages:
90
+ - English - primary and largest test set,
91
+ - German,
92
+ - Spanish,
93
+ - French,
94
+ - Italian.
95
 
96
  ## Dataset Structure
97
 
 
105
 
106
  ### Data Splits
107
 
108
+ The MOCKS testset is split by language, source dataset and OV-KWS type. Each split is divided into:
109
+ - positive examples - test examples with true keyword, 5000-8000 keywords in each subset,
110
+ - similar examples - test examples with similar phrases to keyword selected based on phonetic transcription distance,
111
+ - different examples - test examples with completaly different prases.
112
+
113
+ Each split also contains subset of whole data to allow faster evaluation.
114
 
115
  ## Dataset Creation
116
 
117
+ The MOCKS testset was created from LibriSpeech and Mozilla Common Voice (MCV) datasets that are publicly available. To create it:
118
+ - a [MFA](https://mfa-models.readthedocs.io/en/latest/acoustic/index.html) with publicly available models was used to extract word-level alignments,
119
+ - an internally-developed, rule-based grapheme-to-phoneme (G2P) algorithm was used to prepare phonetic transcriptions for each sample.
120
+
121
+ The data is stored in a 16-bit, single-channel WAV format. 16kHz sampling rate is used for LibriSpeech based testset
122
+ and 48kHz sampling rate for MCV based testset.
123
+
124
+ The offline testset contains additional 0.1 second at the beginning and end of extracted audio sample to mitigate the cut-speech effect.
125
+ The online version contrains additional 1 second or so at the beginning and end of extracted audio sample.
126
+
127
  ### Curation Rationale
128
 
129
  [More Information Needed]
 
160
 
161
  ### Discussion of Biases
162
 
163
+ The MOCKS testset is speaker gender balanced.
164
 
165
  ### Other Known Limitations
166
 
 
178
 
179
  ### Citation Information
180
 
181
+ ```bibtex
182
+ @inproceedings{pudo23_interspeech,
183
+ author={Miko\l{}aj Pudo and Mateusz Wosik and Adam Cie\'slak and Justyna Krzywdziak and Bo\.{z}ena \L{}ukasiak and Artur Janicki},
184
+ title={{MOCKS} 1.0: Multilingual Open Custom Keyword Spotting Testset},
185
+ year={in press.},
186
+ booktitle={Proc. Interspeech 2023},
187
+ }
188
+ ```
189
 
190
  ### Contributions
191