carlosdanielhernandezmena commited on
Commit
1f24384
1 Parent(s): 3f683ed

Last important add to the README file

Browse files
Files changed (1) hide show
  1. README.md +43 -16
README.md CHANGED
@@ -57,16 +57,16 @@ task_ids: []
57
 
58
  ### Dataset Summary
59
  The corpus "RAVNURSSON FAROESE SPEECH AND TRANSCRIPTS" (or RAVNURSSON Corpus for short) is a collection of speech recordings with transcriptions intended for Automatic Speech Recognition (ASR) applications in the language that is spoken at the Faroe Islands (Faroese). It was curated at the Reykjavík University (RU) in 2022.
 
60
  The RAVNURSSON Corpus is an extract of the "Basic Language Resource Kit 1.0" (BLARK 1.0) [1] developed by the Ravnur Project from the Faroe Islands [2]. As a matter of fact, the name RAVNURSSON comes from Ravnur (a tribute to the Ravnur Project) and the suffix "son" which in Icelandic means "son of". Therefore, the name "RAVNURSSON" means "The (Icelandic) son of Ravnur". The double "ss" is just for aesthetics.
 
61
  The audio was collected by recording speakers reading texts. The participants are aged 15-83, divided into 3 age groups: 15-35, 36-60 and 61+.
 
62
  The speech files are made of 249 female speakers and 184 male speakers; 433 speakers total. The recordings were made on TASCAM DR-40 Linear PCM audio recorders using the built-in stereo microphones in WAVE 16 bit with a sample rate of 48kHz, but then, downsampled to 16kHz@16bit mono for this corpus.
63
 
64
- [1] Simonsen, A., Debess, I. N., Lamhauge, S. S., & Henrichsen, P. J. Creating
65
- a basic language resource kit for Faroese. In LREC 2022. 13th International
66
- Conference on Language Resources and Evaluation.
67
 
68
- [2] Website. The Project Ravnur under the Talutøkni Foundation
69
- https://maltokni.fo/en/the-ravnur-project
70
 
71
  ### Example Usage
72
  The RAVNURSSON Corpus is divided in 3 splits: train, validation and test. To load a specific split pass its name as a config name:
@@ -74,42 +74,62 @@ The RAVNURSSON Corpus is divided in 3 splits: train, validation and test. To loa
74
  from datasets import load_dataset
75
  ravnursson = load_dataset("carlosdanielhernandezmena/ravnursson_asr")
76
  ```
77
- To load an specific split (for example, the train split) type:
78
  ```python
79
- ravnursson = load_dataset("carlosdanielhernandezmena/ravnursson_asr",split="train")
 
80
  ```
81
 
82
  ### Supported Tasks
83
- * 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).
84
 
85
  ### Languages
86
  The audio is in Faroese.
87
- The RAVNURSSON Corpus transcriptions have been hand verified. The training subset was balanced for phonetic and dialectal coverage; Test and Dev subsets are gender-balanced. Tabular computer-searchable information is included as well as written documentation.
 
88
  ## Dataset Structure
 
89
  ### Data Instances
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
90
 
91
  ### Data Fields
92
  * `audio_id` (string) - id of audio segment
93
  * `audio` (datasets.Audio) - a dictionary containing the path to the audio, the decoded audio array, and the sampling rate. In non-streaming mode (default), the path points to the locally extracted audio. In streaming mode, the path is the relative path of an audio inside its archive (as files are not downloaded and extracted locally).
94
  * `speaker_id` (string) - id of speaker
95
  * `gender` (string) - gender of speaker (male or female)
96
- * `age` (string) - range of age of the speaker: Younger (15-35), Middle-aged (36-60), Elderly (61+).
97
  * `duration` (float32) - duration of the audio file in seconds.
98
  * `normalized_text` (string) - normalized audio segment transcription
99
  * `dialect` (string) - dialect group, for example "Suðuroy" or "Sandoy".
100
 
101
  ### Data Splits
102
- The speech material has been subdivided into portions for training (train), dev (evaluation) and testing (test). Lengths of every portion are: train = 100h08m, test = 4h30m, dev (evaluation)=4h30m.
103
 
104
- To load an specific portion please see the section "Example Usage".
105
 
106
- The test and validation portions contain the excact same amount of audio each (4.5 hours) and the same number of male (10) and female (10) speakers each.
107
 
108
  ## Dataset Creation
109
 
110
  ### Curation Rationale
111
 
112
- The directory called "speech" contains all the speech files of the corpus. The files in the speech directory are divided in three directories: train, dev and test. The train portion is sub-divided in three types of recordings: RDATA1O, RDATA1OP and RDATA2; this is due the organization of the recordings in the original BLARK 1.0. There, the recordings are divided in Rdata1 and Rdata2.
113
 
114
  One main difference between Rdata1 and Rdata2 is that the reading environment for Rdata2 was controlled by a software called "PushPrompt" which is included in the original BLARK 1.0. Another main difference is that in Rdata1 there are some available transcriptions labeled at the phoneme level. For this reason the audio files in the speech directory of the RAVNURSSON corpus are divided in the folders RDATA1O where "O" is for "Orthographic" and RDATA1OP where "O" is for Orthographic and "P" is for phonetic.
115
 
@@ -126,8 +146,6 @@ The dataset was released with normalized text only at an orthographic level in l
126
 
127
  * The utterances were recorded using a TASCAM DR-40.
128
 
129
- * The audio files in this corpus are distributed in a flac format at 16kHz@16bit mono.
130
-
131
  * Participants self-reported their age group, gender, native language and dialect.
132
 
133
  * Participants are aged between 15 to 83 years.
@@ -153,23 +171,31 @@ This is the first ASR corpus in Faroese.
153
 
154
  ### Discussion of Biases
155
  As the number of reading prompts was limited, the common denominator in the RAVNURSSON corpus is that one prompt is read by more than one speaker. This is relevant because is a common practice in ASR to create a language model using the prompts that are found in the train portion of the corpus. That is not recommended for the RAVNURSSON Corpus as it counts with many prompts shared by all the portions and that will produce an important bias in the language modeling task.
 
156
  In this section we present some statistics about the repeated prompts through all the portions of the corpus.
 
157
  - In the train portion:
158
  * Total number of prompts = 65616
159
  * Number of unique prompts = 38646
160
  There are 26970 repeated prompts in the train portion. In other words, 41.1% of the prompts are repeated.
 
161
  - In the test portion:
162
  * Total number of prompts = 3002
163
  * Number of unique prompts = 2887
164
  There are 115 repeated prompts in the test portion. In other words, 3.83% of the prompts are repeated.
 
165
  - In the dev portion:
166
  * Total number of prompts = 3331
167
  * Number of unique prompts = 3302
168
  There are 29 repeated prompts in the dev portion. In other words, 0.87% of the prompts are repeated.
 
169
  - Considering the corpus as a whole:
170
  * Total number of prompts = 71949
171
  * Number of unique prompts = 39945
172
  There are 32004 repeated prompts in the whole corpus. In other words, 44.48% of the prompts are repeated.
 
 
 
173
  ### Other Known Limitations
174
 
175
  "RAVNURSSON FAROESE SPEECH AND TRANSCRIPTS" by Carlos Daniel Hernández Mena and Annika Simonsen is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) License with the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
@@ -190,4 +216,5 @@ The dataset was created by Annika Simonsen and curated by Carlos Daniel Hernánd
190
  ```
191
  ### Contributions
192
  This project was made possible under the umbrella of the Language Technology Programme for Icelandic 2019-2023. The programme, which is managed and coordinated by Almannarómur, is funded by the Icelandic Ministry of Education, Science and Culture.
 
193
  Special thanks to Dr. Jón Guðnason, professor at Reykjavík University and head of the Language and Voice Lab (LVL) for providing computational resources.
 
57
 
58
  ### Dataset Summary
59
  The corpus "RAVNURSSON FAROESE SPEECH AND TRANSCRIPTS" (or RAVNURSSON Corpus for short) is a collection of speech recordings with transcriptions intended for Automatic Speech Recognition (ASR) applications in the language that is spoken at the Faroe Islands (Faroese). It was curated at the Reykjavík University (RU) in 2022.
60
+
61
  The RAVNURSSON Corpus is an extract of the "Basic Language Resource Kit 1.0" (BLARK 1.0) [1] developed by the Ravnur Project from the Faroe Islands [2]. As a matter of fact, the name RAVNURSSON comes from Ravnur (a tribute to the Ravnur Project) and the suffix "son" which in Icelandic means "son of". Therefore, the name "RAVNURSSON" means "The (Icelandic) son of Ravnur". The double "ss" is just for aesthetics.
62
+
63
  The audio was collected by recording speakers reading texts. The participants are aged 15-83, divided into 3 age groups: 15-35, 36-60 and 61+.
64
+
65
  The speech files are made of 249 female speakers and 184 male speakers; 433 speakers total. The recordings were made on TASCAM DR-40 Linear PCM audio recorders using the built-in stereo microphones in WAVE 16 bit with a sample rate of 48kHz, but then, downsampled to 16kHz@16bit mono for this corpus.
66
 
67
+ [1] Simonsen, A., Debess, I. N., Lamhauge, S. S., & Henrichsen, P. J. Creating a basic language resource kit for Faroese. In LREC 2022. 13th International Conference on Language Resources and Evaluation.
 
 
68
 
69
+ [2] Website. The Project Ravnur under the Talutøkni Foundation https://maltokni.fo/en/the-ravnur-project
 
70
 
71
  ### Example Usage
72
  The RAVNURSSON Corpus is divided in 3 splits: train, validation and test. To load a specific split pass its name as a config name:
 
74
  from datasets import load_dataset
75
  ravnursson = load_dataset("carlosdanielhernandezmena/ravnursson_asr")
76
  ```
77
+ To load an specific split (for example, the validation split) do:
78
  ```python
79
+ from datasets import load_dataset
80
+ ravnursson = load_dataset("carlosdanielhernandezmena/ravnursson_asr",split="validation")
81
  ```
82
 
83
  ### Supported Tasks
84
+ 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).
85
 
86
  ### Languages
87
  The audio is in Faroese.
88
+ The reading prompts for the RAVNURSSON Corpus have been generated by expert linguists. The whole corpus was balanced for phonetic and dialectal coverage; Test and Dev subsets are gender-balanced. Tabular computer-searchable information is included as well as written documentation.
89
+
90
  ## Dataset Structure
91
+
92
  ### Data Instances
93
+ ```python
94
+ {
95
+ 'audio_id': 'KAM06_151121_0101',
96
+ 'audio': {
97
+ 'path': '/home/carlos/.cache/HuggingFace/datasets/downloads/extracted/32b4a757027b72b8d2e25cd9c8be9c7c919cc8d4eb1a9a899e02c11fd6074536/dev/RDATA2/KAM06_151121/KAM06_151121_0101.flac',
98
+ 'array': array([ 0.0010376 , -0.00521851, -0.00393677, ..., 0.00128174,
99
+ 0.00076294, 0.00045776], dtype=float32),
100
+ 'sampling_rate': 16000
101
+ },
102
+ 'speaker_id': 'KAM06_151121',
103
+ 'gender': 'female',
104
+ 'age': '36-60',
105
+ 'duration': 4.863999843597412,
106
+ 'normalized_text': 'endurskin eru týdningarmikil í myrkri',
107
+ 'dialect': 'sandoy'
108
+ }
109
+ ```
110
 
111
  ### Data Fields
112
  * `audio_id` (string) - id of audio segment
113
  * `audio` (datasets.Audio) - a dictionary containing the path to the audio, the decoded audio array, and the sampling rate. In non-streaming mode (default), the path points to the locally extracted audio. In streaming mode, the path is the relative path of an audio inside its archive (as files are not downloaded and extracted locally).
114
  * `speaker_id` (string) - id of speaker
115
  * `gender` (string) - gender of speaker (male or female)
116
+ * `age` (string) - range of age of the speaker: Younger (15-35), Middle-aged (36-60) or Elderly (61+).
117
  * `duration` (float32) - duration of the audio file in seconds.
118
  * `normalized_text` (string) - normalized audio segment transcription
119
  * `dialect` (string) - dialect group, for example "Suðuroy" or "Sandoy".
120
 
121
  ### Data Splits
122
+ The speech material has been subdivided into portions for training (train), development (evaluation) and testing (test). Lengths of each portion are: train = 100h08m, test = 4h30m, dev (evaluation)=4h30m.
123
 
124
+ To load an specific portion please see the above section "Example Usage".
125
 
126
+ The development and test portions have exactly 10 male and 10 female speakers each and both portions have exactly the same size in hours (4.5h each).
127
 
128
  ## Dataset Creation
129
 
130
  ### Curation Rationale
131
 
132
+ The directory called "speech" contains all the speech files of the corpus. The files in the speech directory are divided in three directories: train, dev and test. The train portion is sub-divided in three types of recordings: RDATA1O, RDATA1OP and RDATA2; this is due to the organization of the recordings in the original BLARK 1.0. There, the recordings are divided in Rdata1 and Rdata2.
133
 
134
  One main difference between Rdata1 and Rdata2 is that the reading environment for Rdata2 was controlled by a software called "PushPrompt" which is included in the original BLARK 1.0. Another main difference is that in Rdata1 there are some available transcriptions labeled at the phoneme level. For this reason the audio files in the speech directory of the RAVNURSSON corpus are divided in the folders RDATA1O where "O" is for "Orthographic" and RDATA1OP where "O" is for Orthographic and "P" is for phonetic.
135
 
 
146
 
147
  * The utterances were recorded using a TASCAM DR-40.
148
 
 
 
149
  * Participants self-reported their age group, gender, native language and dialect.
150
 
151
  * Participants are aged between 15 to 83 years.
 
171
 
172
  ### Discussion of Biases
173
  As the number of reading prompts was limited, the common denominator in the RAVNURSSON corpus is that one prompt is read by more than one speaker. This is relevant because is a common practice in ASR to create a language model using the prompts that are found in the train portion of the corpus. That is not recommended for the RAVNURSSON Corpus as it counts with many prompts shared by all the portions and that will produce an important bias in the language modeling task.
174
+
175
  In this section we present some statistics about the repeated prompts through all the portions of the corpus.
176
+
177
  - In the train portion:
178
  * Total number of prompts = 65616
179
  * Number of unique prompts = 38646
180
  There are 26970 repeated prompts in the train portion. In other words, 41.1% of the prompts are repeated.
181
+
182
  - In the test portion:
183
  * Total number of prompts = 3002
184
  * Number of unique prompts = 2887
185
  There are 115 repeated prompts in the test portion. In other words, 3.83% of the prompts are repeated.
186
+
187
  - In the dev portion:
188
  * Total number of prompts = 3331
189
  * Number of unique prompts = 3302
190
  There are 29 repeated prompts in the dev portion. In other words, 0.87% of the prompts are repeated.
191
+
192
  - Considering the corpus as a whole:
193
  * Total number of prompts = 71949
194
  * Number of unique prompts = 39945
195
  There are 32004 repeated prompts in the whole corpus. In other words, 44.48% of the prompts are repeated.
196
+
197
+ NOTICE!: It is also important to clarify that any of the 3 portions of the corpus share speakers.
198
+
199
  ### Other Known Limitations
200
 
201
  "RAVNURSSON FAROESE SPEECH AND TRANSCRIPTS" by Carlos Daniel Hernández Mena and Annika Simonsen is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) License with the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
 
216
  ```
217
  ### Contributions
218
  This project was made possible under the umbrella of the Language Technology Programme for Icelandic 2019-2023. The programme, which is managed and coordinated by Almannarómur, is funded by the Icelandic Ministry of Education, Science and Culture.
219
+
220
  Special thanks to Dr. Jón Guðnason, professor at Reykjavík University and head of the Language and Voice Lab (LVL) for providing computational resources.