remg1997 commited on
Commit
235ce1a
1 Parent(s): b26f165

Fix readme

Browse files
README.md CHANGED
@@ -12,7 +12,6 @@ task_ids:
12
  - audio-language-identification
13
  viewer: false
14
 
15
-
16
  ---
17
 
18
 
@@ -23,11 +22,9 @@ viewer: false
23
  - [Table of Contents](#table-of-contents)
24
  - [Dataset Description](#dataset-description)
25
  - [Dataset Summary](#dataset-summary)
 
26
  - [Dataset Structure](#dataset-structure)
27
- - [reformat](#reformat)
28
- - [audio and audio2](#transcription-and-transcription_2)
29
- - [license.json](#licensejson)
30
- - [Data License](#data-license)
31
  - [Dataset Creation](#dataset-creation)
32
  - [Source Data](#source-data)
33
  - [Initial Data Collection and Normalization](#initial-data-collection-and-normalization)
@@ -45,28 +42,27 @@ viewer: false
45
 
46
  ### Dataset Summary
47
 
48
- The Unuspervised Peoples Speech Dataset is a compilation of audiofiles with transcriptions extracted from wikimedia commons that is licensed for academic and commercial usage under CC and Public domain. It includes {}+ hours of transcribed speech in different languages with a diverse set of speakers.
49
 
50
- ## Dataset Structure
51
 
52
- ### audio and audio2
53
 
54
- Folders with the raw audio. We split this into two directories because Hugging Face does not support more than 10,000 files in a single directory.
55
 
 
56
 
57
- ### license.json
58
 
59
- File with license information. The key is the name of the original audio file on Wikimedia Commons.
60
 
61
- ### Data License
62
 
63
- Here is an excerpt from license.json:
64
 
65
- """
66
- '"Berlin Wall" Speech - President Reagan\'s Address at the Brandenburg Gate - 6-12-87.webm': {'author': '<td>\n<a class="external text" href="https://www.youtube.com/user/ReaganFoundation" rel="nofollow">ReaganFoundation</a></td>',
67
- 'source': '<td>\n<bdo dir="ltr" lang="en"><a href="/wiki/Commons:YouTube_files" [...],
68
- 'html_license': '[\'<table class="layouttemplate mw-content-ltr" lang="en" style="width:100%; [...],
69
- 'license': 'Public Domain'},
70
 
71
 
72
  ## Dataset Creation
@@ -75,7 +71,7 @@ Here is an excerpt from license.json:
75
 
76
  #### Initial Data Collection and Normalization
77
 
78
- Data was downloaded from
79
 
80
  #### Preprocessing
81
 
@@ -85,7 +81,7 @@ No preprocessing was done.
85
 
86
  #### Annotation process
87
 
88
- No manual annotation is done. We download only source audio with already existing transcripts.
89
 
90
  In particular, there is no "forced alignment" or "segmentation" done on this dataset.
91
 
@@ -97,16 +93,14 @@ Several of our sources are legal and government proceedings, spoken stories, spe
97
 
98
  ### Discussion of Biases
99
 
100
- Our data is downloaded from commons.wikimedia.org. As such, the data is biased towards whatever users decide to upload there.
101
 
102
- The data is also mostly English, though this has potential for multitask learning because several audio files have more than one transcription.
103
 
104
  ## Additional Information
105
 
106
  ### Licensing Information
107
 
108
- The source data contains data under Public Domain and Creative Commons Licenses.
109
 
110
  We license this dataset under https://creativecommons.org/licenses/by-sa/4.0/
111
-
112
- The appropriate attributions are in license.json
 
12
  - audio-language-identification
13
  viewer: false
14
 
 
15
  ---
16
 
17
 
 
22
  - [Table of Contents](#table-of-contents)
23
  - [Dataset Description](#dataset-description)
24
  - [Dataset Summary](#dataset-summary)
25
+ - [Relevant Statistics](#relevant-statistics)
26
  - [Dataset Structure](#dataset-structure)
27
+ - [Audio folders](#audio-folders)
 
 
 
28
  - [Dataset Creation](#dataset-creation)
29
  - [Source Data](#source-data)
30
  - [Initial Data Collection and Normalization](#initial-data-collection-and-normalization)
 
42
 
43
  ### Dataset Summary
44
 
45
+ The Unsupervised Peoples Speech Dataset is a compilation of audiofiles extracted from Archive.org that is licensed for academic and commercial usage under CC-BY and CC-BY-SA licenses. It includes more than one million hours of audio with a diverse set of speakers.
46
 
47
+ ### Relevant Statistics
48
 
49
+ #### Duration Distribution
50
 
51
+ Most of the audios range between 1 and 10 minutes in length, with only 14 of them exceeding the 100 hour mark.
52
 
53
+ ![Duration Distribution](./images/duration_distribution.png)
54
 
55
+ #### Sample Rates
56
 
57
+ 99% of the audio in the dataset has a 44.1Khz sample rate, and the remaining audio varies from the more common 16Khz, 24Khz and 48 Khz to custom sample rates.
58
 
59
+ ![Sample Rates](./images/sample_rate_distribution.png)
60
 
61
+ ## Dataset Structure
62
 
63
+ ### Audio folders
64
+
65
+ Folders with the raw audio. We split this into two directories because Hugging Face does not support more than 10,000 files in a single directory.
 
 
66
 
67
 
68
  ## Dataset Creation
 
71
 
72
  #### Initial Data Collection and Normalization
73
 
74
+ Data was downloaded via the archive.org API. No data inference was done.
75
 
76
  #### Preprocessing
77
 
 
81
 
82
  #### Annotation process
83
 
84
+ No manual annotation is done. We download only source audio.
85
 
86
  In particular, there is no "forced alignment" or "segmentation" done on this dataset.
87
 
 
93
 
94
  ### Discussion of Biases
95
 
96
+ Our data is downloaded from archive.org. As such, the data is biased towards whatever users decide to upload there.
97
 
98
+ Almost all of our data is American accented English.
99
 
100
  ## Additional Information
101
 
102
  ### Licensing Information
103
 
104
+ The source data contains data under CC-BY-SA and CC-BY licenses.
105
 
106
  We license this dataset under https://creativecommons.org/licenses/by-sa/4.0/
 
 
images/duration_distribution.png ADDED

Git LFS Details

  • SHA256: f1bfe678963dbe9d2e0809f76024ef48081fe7ddd58df2c4612e53d82543b692
  • Pointer size: 130 Bytes
  • Size of remote file: 16.8 kB
images/sample_rate_distribution.png ADDED

Git LFS Details

  • SHA256: f88b26674cd5b746104237f815942856011bbfa582f0477b12831ef26724b210
  • Pointer size: 130 Bytes
  • Size of remote file: 22.4 kB