File size: 6,146 Bytes
685fe60
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
877d9bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1fe9378
 
 
 
 
877d9bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: speaker_id
    dtype: int64
  - name: chapter_id
    dtype: int64
  - name: path
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 2818850887.0
    num_examples: 22263
  - name: test
    num_bytes: 59879514.0
    num_examples: 457
  download_size: 2877012819
  dataset_size: 2878730401.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---


# Zeroth-Korean Dataset

## Introduction
The Zeroth-Korean dataset is a publicly available speech dataset created for Korean automatic speech recognition (ASR) research and development. This dataset is distributed under the CC BY 4.0 license, allowing anyone to use it freely. The goal of the Zeroth project is to make Korean speech recognition more widely accessible.

## Dataset Overview
- **Total Data**: Approximately 51.6 hours of training data and 1.2 hours of test data
- **Number of Utterances**: 22,263 utterances in the training set and 457 utterances in the test set
- **Number of Speakers**: 105 speakers in the training set, 10 speakers in the test set
- **Sampling Rate**: 16kHz

## Version Information
### Version 2.0.0
- **Changes**:
  - Instead of downloading the original files directly from openslr.org, the data is distributed using pyarrow for more efficient storage.
  - The `audio` field has been updated from `numpy.ndarray` to `datasets.Audio` type for improved accessibility of audio data.
  - The `file` field has been renamed to `path`.

## Dataset Composition
- **Training Data (Train)**: Consists of 51.6 hours of Korean utterances, with a total of 22,263 utterances. This data is used to train speech recognition models.
- **Test Data (Test)**: Consists of 1.2 hours of utterances, with a total of 457 utterances, used for evaluating model performance.

### Data Fields
- **id**: Unique identifier for each utterance (string)
- **speaker_id**: Speaker identifier (int64)
- **chapter_id**: Chapter identifier (int64)
- **path**: Path to the audio file (string)
- **audio**: Audio data (managed as `datasets.Audio` type, sampling rate 16kHz)
- **text**: Transcribed text of the utterance (string)

## Usage Example
The Zeroth-Korean dataset can be easily loaded using the Hugging Face library.

```python
from datasets import load_dataset

# Load Zeroth-Korean dataset
dataset = load_dataset("kresnik/zeroth_korean")

# Print an example from the training dataset
print(dataset['train'][0])
```

## License
This dataset follows the [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) license. This license allows the data to be used freely, provided the source is properly credited.

The uploader has transferred the publicly available data from openslr.org to the Hugging Face Hub and did not participate in the creation of the dataset. This dataset is provided to facilitate broader accessibility of the original dataset.

The Zeroth project was developed through the collaboration of Lucas Jo (@Atlas Guide Inc.) and Wonkyum Lee (@Gridspace Inc.).
- **Contact**: Lucas Jo (lucasjo@goodatlas.com), Wonkyum Lee (wonkyum@gridspace.com)







# Zeroth-Korean ๋ฐ์ดํ„ฐ์…‹ (ํ•œ๊ธ€)

## ์†Œ๊ฐœ
Zeroth-Korean ๋ฐ์ดํ„ฐ์…‹์€ ํ•œ๊ตญ์–ด ์Œ์„ฑ ์ธ์‹ ์—ฐ๊ตฌ์™€ ๊ฐœ๋ฐœ์„ ์œ„ํ•ด ์ œ์ž‘๋œ ๊ณต๊ฐœ ์Œ์„ฑ ๋ฐ์ดํ„ฐ์…‹์ž…๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ์…‹์€ CC BY 4.0 ๋ผ์ด์„ ์Šค ํ•˜์— ๋ฐฐํฌ๋˜๋ฉฐ, ๋ˆ„๊ตฌ๋‚˜ ์ž์œ ๋กญ๊ฒŒ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. Zeroth ํ”„๋กœ์ ํŠธ์˜ ๋ชฉํ‘œ๋Š” ํ•œ๊ตญ์–ด ์Œ์„ฑ ์ธ์‹์„ ๋” ๋„๋ฆฌ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

## ๋ฐ์ดํ„ฐ์…‹ ๊ฐœ์š”
- **์ด ๋ฐ์ดํ„ฐ๋Ÿ‰**: ์•ฝ 51.6์‹œ๊ฐ„์˜ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์™€ 1.2์‹œ๊ฐ„์˜ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ
- **๋ฐœํ™” ์ˆ˜**: 22,263๊ฐœ์˜ ํ›ˆ๋ จ ๋ฐœํ™”์™€ 457๊ฐœ์˜ ํ…Œ์ŠคํŠธ ๋ฐœํ™”
- **ํ™”์ž ์ˆ˜**: ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์—์„œ 105๋ช…์˜ ํ™”์ž, ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ์—์„œ 10๋ช…์˜ ํ™”์ž
- **์ƒ˜ํ”Œ๋ง ๋ ˆ์ดํŠธ**: 16kHz

## ๋ฒ„์ „ ์ •๋ณด
### Version 2.0.0
- **๋ณ€๊ฒฝ ์‚ฌํ•ญ**:
  - openslr.org์—์„œ ์›๋ณธ ํŒŒ์ผ์„ ์ง์ ‘ ๋‹ค์šด๋กœ๋“œํ•˜๋Š” ๋ฐฉ์‹ ๋Œ€์‹  pyarrow๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ถ„์‚ฐ ์ €์žฅ
  - `audio` ํ•„๋“œ๋ฅผ `numpy.ndarray` ๋Œ€์‹  `datasets.Audio` ํƒ€์ž…์œผ๋กœ ์—…๋ฐ์ดํŠธํ•˜์—ฌ ์˜ค๋””์˜ค ๋ฐ์ดํ„ฐ ์ ‘๊ทผ์„ฑ ํ–ฅ์ƒ
  - ๊ธฐ์กด `file` ํ•„๋“œ๋ฅผ `path`๋กœ ์ด๋ฆ„ ๋ณ€๊ฒฝ

## ๋ฐ์ดํ„ฐ ๊ตฌ์„ฑ
- **ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ (Train)**: 51.6์‹œ๊ฐ„์˜ ํ•œ๊ตญ์–ด ๋ฐœํ™” ๋ฐ์ดํ„ฐ๋กœ ๊ตฌ์„ฑ๋˜๋ฉฐ, ๋ฐœํ™” ์ˆ˜๋Š” ์ด 22,263๊ฐœ์ž…๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ๋Š” ์Œ์„ฑ ์ธ์‹ ๋ชจ๋ธ์˜ ํ›ˆ๋ จ์— ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.
- **ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ (Test)**: 1.2์‹œ๊ฐ„์˜ ๋ฐœํ™” ๋ฐ์ดํ„ฐ๋กœ, ์ด 457๊ฐœ์˜ ๋ฐœํ™”๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๋ชจ๋ธ ์„ฑ๋Šฅ ๊ฒ€์ฆ์— ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.

### ๋ฐ์ดํ„ฐ ํ•„๋“œ
- **id**: ๋ฐœํ™” ๊ณ ์œ  ์‹๋ณ„์ž (string)
- **speaker_id**: ํ™”์ž ์‹๋ณ„ ๋ฒˆํ˜ธ (int64)
- **chapter_id**: ์ฑ•ํ„ฐ ์‹๋ณ„ ๋ฒˆํ˜ธ (int64)
- **path**: ์˜ค๋””์˜ค ํŒŒ์ผ ๊ฒฝ๋กœ (string)
- **audio**: ์˜ค๋””์˜ค ๋ฐ์ดํ„ฐ (`datasets.Audio` ํƒ€์ž…์œผ๋กœ ๊ด€๋ฆฌ, ์ƒ˜ํ”Œ๋ง ๋ ˆ์ดํŠธ 16kHz)
- **text**: ๋ฐœํ™”๋œ ๋ฌธ์žฅ (string)

## ์‚ฌ์šฉ ์˜ˆ์‹œ
Hugging Face ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ๋ฅผ ํ†ตํ•ด Zeroth-Korean ๋ฐ์ดํ„ฐ์…‹์„ ์‰ฝ๊ฒŒ ๋กœ๋“œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

```python
from datasets import load_dataset

# Zeroth-Korean ๋ฐ์ดํ„ฐ์…‹ ๋กœ๋“œ
dataset = load_dataset("kresnik/zeroth_korean")

# ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์…‹ ์˜ˆ์‹œ ์ถœ๋ ฅ
print(dataset['train'][0])
```

## ๋ผ์ด์„ ์Šค
์ด ๋ฐ์ดํ„ฐ์…‹์€ [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) ๋ผ์ด์„ ์Šค๋ฅผ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค. ์ด ๋ผ์ด์„ ์Šค๋Š” ๋ฐ์ดํ„ฐ๋ฅผ ์ž์œ ๋กญ๊ฒŒ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ํ—ˆ์šฉํ•˜๋˜, ์ถœ์ฒ˜๋ฅผ ๋ฐ˜๋“œ์‹œ ๋ช…์‹œํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

๊ฒŒ์‹œ์ž๋Š” openslr.org์— ๊ณต๊ฐœ๋œ ๋ฐ์ดํ„ฐ๋ฅผ Hugging Face Hub์— ์˜ฎ๊ฒจ ๋ฐฐํฌํ•˜๋Š” ์—ญํ• ์„ ์ˆ˜ํ–‰ํ•˜์˜€์œผ๋ฉฐ, ๋ฐ์ดํ„ฐ์…‹ ์ œ์ž‘์—๋Š” ์ฐธ์—ฌํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ์…‹์€ ์›๋ณธ ๋ฐ์ดํ„ฐ์…‹์˜ ๋ฐฐํฌ๋ฅผ ๋•๊ธฐ ์œ„ํ•œ ๋ชฉ์ ์œผ๋กœ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค.

Zeroth ํ”„๋กœ์ ํŠธ๋Š” Lucas Jo (@Atlas Guide Inc.)์™€ Wonkyum Lee (@Gridspace Inc.)์˜ ํ˜‘๋ ฅ์œผ๋กœ ๊ฐœ๋ฐœ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
- **์—ฐ๋ฝ์ฒ˜**: Lucas Jo (lucasjo@goodatlas.com), Wonkyum Lee (wonkyum@gridspace.com)