File size: 7,505 Bytes
3997156 2acc6f6 3997156 0c387d3 3997156 0c387d3 3997156 0c387d3 3997156 af7f7b7 4e7ffcb af7f7b7 4e7ffcb af7f7b7 4e7ffcb af7f7b7 4e7ffcb af7f7b7 4e7ffcb 306b856 4e7ffcb 306b856 4e7ffcb 306b856 4e7ffcb 306b856 4e7ffcb 306b856 4e7ffcb 306b856 4e7ffcb f9ab549 4e7ffcb f9ab549 4e7ffcb f9ab549 4e7ffcb 306b856 f9ab549 306b856 4e7ffcb f9ab549 4e7ffcb 306b856 f9ab549 306b856 f9ab549 306b856 f9ab549 306b856 f9ab549 306b856 f9ab549 306b856 4e7ffcb f9ab549 306b856 f9ab549 4e7ffcb 306b856 af7f7b7 306b856 4e7ffcb af7f7b7 4e7ffcb af7f7b7 4e7ffcb af7f7b7 eb4e194 af7f7b7 306b856 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 |
---
language:
- da
license: openrail
size_categories:
- 100K<n<1M
task_categories:
- automatic-speech-recognition
- audio-classification
pretty_name: CoRal
dataset_info:
config_name: read_aloud
features:
- name: id_recording
dtype: string
- name: id_sentence
dtype: string
- name: id_speaker
dtype: string
- name: text
dtype: string
- name: location
dtype: string
- name: location_roomdim
dtype: string
- name: noise_level
dtype: string
- name: noise_type
dtype: string
- name: source_url
dtype: string
- name: age
dtype: int64
- name: gender
dtype: string
- name: dialect
dtype: string
- name: country_birth
dtype: string
- name: validated
dtype: string
- name: audio
dtype: audio
- name: asr_prediction
dtype: string
- name: asr_validation_model
dtype: string
- name: asr_wer
dtype: float64
- name: asr_cer
dtype: float64
splits:
- name: train
num_bytes: 125338742145.4203
num_examples: 231292
- name: val
num_bytes: 997108786.9341497
num_examples: 1840
- name: test
num_bytes: 3229765418.5475717
num_examples: 5960
download_size: 121228528051
dataset_size: 129565616350.90202
configs:
- config_name: read_aloud
data_files:
- split: train
path: read_aloud/train-*
- split: val
path: read_aloud/val-*
- split: test
path: read_aloud/test-*
---
# CoRal: Danish Conversational and Read-aloud Dataset
## Dataset Overview
**CoRal** is a comprehensive Automatic Speech Recognition (ASR) dataset designed to
capture the diversity of the Danish language across various dialects, accents, genders,
and age groups. The primary goal of the CoRal dataset is to provide a robust resource
for training and evaluating ASR models that can understand and transcribe spoken Danish
in all its variations.
### Key Features
- **Dialect and Accent Diversity**: The dataset includes speech samples from all major
Danish dialects as well as multiple accents, ensuring broad geographical coverage and
the inclusion of regional linguistic features.
- **Gender Representation**: Both male and female speakers are well-represented,
offering balanced gender diversity.
- **Age Range**: The dataset includes speakers from a wide range of age groups,
providing a comprehensive resource for age-agnostic ASR model development.
- **High-Quality Audio**: All recordings are of high quality, ensuring that the dataset
can be used for both training and evaluation of high-performance ASR models.
### Quick Start
The CoRal dataset is ideal for training ASR models that need to generalise across
different dialects and speaker demographics within the Danish language. Below is an
example of how to load and use the dataset with Hugging Face's `datasets` library:
```python
from datasets import load_dataset
# Load the Coral dataset
coral = load_dataset("alexandrainst/coral", "read_aloud")
# Example: Accessing an audio sample and its transcription
sample = coral['train'][0]
audio = sample['audio']
transcription = sample['text']
print(f"Audio: {audio['array']}")
print(f"Text: {transcription}")
```
## Data Fields
- `id_recording`: Unique identifier for the recording.
- `id_sentence`: Unique identifier for the text being read aloud.
- `id_speaker`: Unique identifier for each speaker.
- `text`: Text being read aloud.
- `location`: Address of recording place.
- `location_roomdim`: Dimension of recording room.
- `noise_level`: Noise level in the room, given in dB.
- `noise_type`: Noise exposed to the speaker while recording. Note the noise is not
present in the audio, but is there to mimic differences in speech in a noisy
environment.
- `source_url`: URL to the source of the text.
- `age`: Age of the speaker.
- `gender`: Gender of the speaker.
- `dialect`: Self-reported dialect of the speaker.
- `country_birth`: Country where the speaker was born.
- `validated`: Manual validation state of the recording.
- `audio`: The audio file of the recording.
- `asr_prediction`: ASR output prediction of the `asr_validation_model`.
- `asr_validation_model`: Hugging Face Model ID used for automatic validation of the
recordings.
- `asr_wer`: Word error rate between `asr_prediction` and `text`.
- `asr_cer`: Character error rate between `asr_prediction` and `text`.
## Read-aloud Data Statistics
### Train Split
There are ~365 hours of audio in the train split, with ~565 speakers, reading ~152,000
unique sentences aloud.
Gender distribution:
- female: 74%
- male: 26%
Dialect distribution:
- Bornholmsk: 2%
- Fynsk: 5%
- Københavnsk: 15%
- Nordjysk: 16%
- Sjællandsk: 14%
- Sydømål: 0.2%
- Sønderjysk: 5%
- Vestjysk: 11%
- Østjysk: 33%
Accent distribution:
- native: 97%
- foreign: 3%
Age group distribution:
- 0-24: 6%
- 25-49: 36%
- 50-: 57%
### Validation Split
There are 3.05 hours of audio in the validation split, with 11 speakers.
Gender distribution:
- female: 58%
- male: 42%
Dialect distribution:
- Bornholmsk: 19%
- Fynsk: 10%
- Københavnsk: 17%
- Nordjysk: 5%
- Sjællandsk: 11%
- Sydømål: 13%
- Sønderjysk: 11%
- Vestjysk: 7%
- Østjysk: 8%
Accent distribution:
- native: 68%
- foreign: 32%
Age group distribution:
- 0-24: 29%
- 25-49: 37%
- 50-: 35%
### Test Split
There are 9.43 hours of audio in the test split, with 23 speakers.
Gender distribution:
- female: 55%
- male: 45%
Dialect distribution:
- Bornholmsk: 10%
- Fynsk: 11%
- Københavnsk: 10%
- Nordjysk: 12%
- Sjællandsk: 12%
- Sydømål: 11%
- Sønderjysk: 10%
- Vestjysk: 10%
- Østjysk: 13%
Accent distribution:
- native: 88%
- foreign: 12%
Age group distribution:
- 0-24: 29%
- 25-49: 42%
- 50-: 29%
## Conversational Data Statistics
The conversational data is not yet available, but we are working on it and plan to
release it during 2024.
## Example Use Cases
### ASR Model Training
Train robust ASR models that can handle dialectal variations and diverse speaker
demographics in Danish.
### Dialect Studies
Analyse the linguistic features of different Danish dialects.
### Forbidden Use Cases
Speech Synthesis and Biometric Identification are not allowed using the CoRal dataset.
For more information, see addition 4 in our
[license](https://huggingface.co/datasets/alexandrainst/coral/blob/main/LICENSE).
## License
The dataset is licensed under a custom license, adapted from OpenRAIL-M, which allows
commercial use with a few restrictions (speech synthesis and biometric identification).
See [license](https://huggingface.co/datasets/alexandrainst/coral/blob/main/LICENSE).
## Creators and Funders
The CoRal project is funded by the [Danish Innovation
Fund](https://innovationsfonden.dk/) and consists of the following partners:
- [Alexandra Institute](https://alexandra.dk/)
- [University of Copenhagen](https://www.ku.dk/)
- [Agency for Digital Government](https://digst.dk/)
- [Alvenir](https://www.alvenir.ai/)
- [Corti](https://www.corti.ai/)
## Citation
We will submit a research paper soon, but until then, if you use the CoRal dataset in
your research or development, please cite it as follows:
```bibtex
@dataset{coral2024,
author = {Dan Saattrup Nielsen, Sif Bernstorff Lehmann, Simon Leminen Madsen, Anders Jess Pedersen, Anna Katrine van Zee and Torben Blach},
title = {CoRal: A Diverse Danish ASR Dataset Covering Dialects, Accents, Genders, and Age Groups},
year = {2024},
url = {https://hf.co/datasets/alexandrainst/coral},
}
``` |