rryner commited on
Commit
3ed5a41
·
verified ·
1 Parent(s): c8305d6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +94 -88
README.md CHANGED
@@ -1,88 +1,94 @@
1
- ---
2
- license: cc-by-nc-4.0
3
- dataset_info:
4
- features:
5
- - name: audio
6
- dtype:
7
- audio:
8
- sampling_rate: 44100
9
- - name: question
10
- dtype: string
11
- - name: option1
12
- dtype: string
13
- - name: option2
14
- dtype: string
15
- - name: option3
16
- dtype: string
17
- - name: option4
18
- dtype: string
19
- - name: target
20
- dtype: string
21
- download_size: 10988605093
22
- dataset_size: 11198533410
23
- tags:
24
- - audio
25
- - in-the-wild
26
- - compound-task
27
- language:
28
- - en
29
- - zh
30
- - ru
31
- - th
32
- multilinguality:
33
- - multilingual
34
- pretty_name: GlobeAudio
35
- configs:
36
- - config_name: en
37
- data_files:
38
- - split: train
39
- path: data/en/en-*
40
- default: true
41
- - config_name: zh
42
- data_files:
43
- - split: train
44
- path: data/zh/zh-*
45
- - config_name: sg
46
- data_files:
47
- - split: train
48
- path: data/sg/sg-*
49
- - config_name: ru
50
- data_files:
51
- - split: train
52
- path: data/ru/ru-*
53
- - config_name: th
54
- data_files:
55
- - split: train
56
- path: data/th/th-*
57
- ---
58
-
59
- # GlobeAudio Dataset
60
-
61
- ## Table of Contents
62
- - [Dataset Description](#dataset-description)
63
- - [Language Selection](#language-selection)
64
- - [Dataset Quality](#dataset-quality)
65
- - [Dataset Breakdown](#dataset-breakdown)
66
-
67
-
68
- ## Dataset Description
69
- GlobeAudio is a multilingual, multicultural in-the-Wild benchmark for assessing compound audio understanding, comprising of 4,918 carefully constructed MCQs across five typologically diverse languages, namely English (United States), Chinese (China), Thai (Thailand), Russian (Russia) and Singlish (Singapore). The dataset comprises of naturally occurring audio clips from online media that reflect real-life speech across diverse contexts, and are curated to rigorously examine the ability of multilingual systems to handle acoustic and sociolinguistic variability.
70
-
71
- ## Language Selection
72
- The 5 carefully selected languages span multiple language families (Indo-European, Sino-Tibetan, Tai-Kadai and creole varieties), covers a wide range of geographic regions, and reflects substantial variation in resource availability. They also differ markedly in script systems and spoken characteristics -- the inclusion of Singlish enables the evaluation of naturally occuring code-mixing and multilingual speech within a single variety, capturing linguistic phenomena that are common in real-world audio but rarely represented in existing benchmarks.
73
-
74
- ## Dataset Quality
75
- For all selected languages, data annotation is performed exclusively by native speakers for whom the language is their first language, in order to ensure high linguistic fidelity, culturally appropriate interpretation and reliable construction of questions and distractors. Consequently, a two-stage quality control process was conducted involving random sampling and cross-checking, producing a final dataset with an inter-annotator agreement of 95.5%, together with verified and consistent gold labels.
76
-
77
- ## Dataset Breakdown
78
- Language | Code | Country | Family | CC Size | Clip length | Q/A length | Total
79
- :------ | :----- | :----- | :----- | :----- | :-------- | :-------- | :-------:
80
- English | en | USA | Indo-European | 41.41 | 25.72 | 53.80 / 22.89 | 1,274
81
- Russian | ru | Russia | Indo-European | 6.48 | 25.69 | 37.83 / 22.19 | 924
82
- Chinese | zh | China | Sino-Tibetan | 6.18 | 24.88 | 13.53 / 5.01 | 1,072
83
- Thai | th | Thailand | Tai-Kadai | 0.39 | 24.82 | 29.26 / 12.98 | 1,145
84
- Singlish | sg | Singapore | - (Variety) | - | 25.54 | 73.29 / 34.92 | 503
85
- Total | | | | | 25.30 | 38.30 / 17.79 | 4,918
86
-
87
- Each audio clip is approximately 20-40 seconds long, and contains naturally occurring audio that reflect real-life auditory settings spanning both formal and informal contexts. The mean and median clip length are 25.33s and 24.95s respectively.
88
-
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-4.0
3
+ dataset_info:
4
+ features:
5
+ - name: audio
6
+ dtype:
7
+ audio:
8
+ sampling_rate: 44100
9
+ - name: question
10
+ dtype: string
11
+ - name: option1
12
+ dtype: string
13
+ - name: option2
14
+ dtype: string
15
+ - name: option3
16
+ dtype: string
17
+ - name: option4
18
+ dtype: string
19
+ - name: target
20
+ dtype: string
21
+ download_size: 10988605093
22
+ dataset_size: 11198533410
23
+ tags:
24
+ - audio
25
+ - in-the-wild
26
+ - compound-task
27
+ language:
28
+ - en
29
+ - zh
30
+ - ru
31
+ - th
32
+ multilinguality:
33
+ - multilingual
34
+ pretty_name: GlobeAudio
35
+ configs:
36
+ - config_name: en
37
+ data_files:
38
+ - split: train
39
+ path: data/en/en-*
40
+ default: true
41
+ - config_name: zh
42
+ data_files:
43
+ - split: train
44
+ path: data/zh/zh-*
45
+ - config_name: sg
46
+ data_files:
47
+ - split: train
48
+ path: data/sg/sg-*
49
+ - config_name: ru
50
+ data_files:
51
+ - split: train
52
+ path: data/ru/ru-*
53
+ - config_name: th
54
+ data_files:
55
+ - split: train
56
+ path: data/th/th-*
57
+ ---
58
+
59
+ # GlobeAudio Dataset
60
+
61
+ ## Table of Contents
62
+ - [Dataset Description](#dataset-description)
63
+ - [Language Selection](#language-selection)
64
+ - [Dataset Quality](#dataset-quality)
65
+ - [Dataset Breakdown](#dataset-breakdown)
66
+ - [Sample Usage](#sample-usage)
67
+
68
+
69
+ ## Dataset Description
70
+ GlobeAudio is a multilingual, multicultural in-the-Wild benchmark for assessing compound audio understanding, comprising of 4,918 carefully constructed MCQs across five typologically diverse languages, namely English (United States), Chinese (China), Thai (Thailand), Russian (Russia) and Singlish (Singapore). The dataset comprises of naturally occurring audio clips from online media that reflect real-life speech across diverse contexts, and are curated to rigorously examine the ability of multilingual systems to handle acoustic and sociolinguistic variability.
71
+
72
+ ## Language Selection
73
+ The 5 carefully selected languages span multiple language families (Indo-European, Sino-Tibetan, Tai-Kadai and creole varieties), covers a wide range of geographic regions, and reflects substantial variation in resource availability. They also differ markedly in script systems and spoken characteristics -- the inclusion of Singlish enables the evaluation of naturally occuring code-mixing and multilingual speech within a single variety, capturing linguistic phenomena that are common in real-world audio but rarely represented in existing benchmarks.
74
+
75
+ ## Dataset Quality
76
+ For all selected languages, data annotation is performed exclusively by native speakers for whom the language is their first language, in order to ensure high linguistic fidelity, culturally appropriate interpretation and reliable construction of questions and distractors. Consequently, a two-stage quality control process was conducted involving random sampling and cross-checking, producing a final dataset with an inter-annotator agreement of 95.5%, together with verified and consistent gold labels.
77
+
78
+ ## Dataset Breakdown
79
+ Language | Code | Country | Family | CC Size | Clip length | Q/A length | Total
80
+ :------ | :----- | :----- | :----- | :----- | :-------- | :-------- | :-------:
81
+ English | en | USA | Indo-European | 41.41 | 25.72 | 53.80 / 22.89 | 1,274
82
+ Russian | ru | Russia | Indo-European | 6.48 | 25.69 | 37.83 / 22.19 | 924
83
+ Chinese | zh | China | Sino-Tibetan | 6.18 | 24.88 | 13.53 / 5.01 | 1,072
84
+ Thai | th | Thailand | Tai-Kadai | 0.39 | 24.82 | 29.26 / 12.98 | 1,145
85
+ Singlish | sg | Singapore | - (Variety) | - | 25.54 | 73.29 / 34.92 | 503
86
+ Total | | | | | 25.30 | 38.30 / 17.79 | 4,918
87
+
88
+ Each audio clip is approximately 20-40 seconds long, and contains naturally occurring audio that reflect real-life auditory settings spanning both formal and informal contexts. The mean and median clip length are 25.33s and 24.95s respectively.
89
+
90
+ ## Sample Usage
91
+ ```
92
+ from datasets import load_dataset
93
+ dataset = load_dataset("iNLP-Lab/GlobeAudio")
94
+ ```