Dataset · Bingsu
Korean speech corpus with transcriptions
Bingsu/zeroth-korean
About 53 hours of Korean speech with text transcriptions for training and evaluating ASR models.
What's inside
The dataset contains Korean audio recordings and their text transcriptions. The training set includes 51.6 hours of speech (22,263 utterances from 105 speakers across 3,000 sentences), and the test set includes 1.2 hours (457 utterances from 10 speakers). The corpus also includes a pretrained language model, lexicon, and the Morfessor morpheme segmenter.
What you can do with it
- Training and fine-tuning ASR models for Korean speech recognition
- Testing Korean speech recognition quality on a benchmark split
- Tasks
- Automatic speech recognition
- Languages
- Korean
- Size
- 10K–100K rows
- Downloads this month
- 1.4K
- Likes
- 46
Models trained on it
- mradermacher/Thaki-ASR-KO-Dialects-v1-i1-GGUF 3.3K
- ghost613/whisper-large-v3-turbo-koreanAutomatic speech recognition 1.4K
- mradermacher/Thaki-ASR-KO-1.7B-GGUF 930
- mradermacher/Thaki-ASR-KO-Dialects-v1-GGUF 751
- ghost613/faster-whisper-large-v3-turbo-koreanAutomatic speech recognition 508
- litert-community/SenseVoiceSmall-ko-zerothAutomatic speech recognition 193
- ThakiCloud/Thaki-ASR-KO-1.7BAutomatic speech recognition 163
- ThakiCloud/Thaki-ASR-KO-Dialects-v1Automatic speech recognition 132
How to load it
from datasets import load_dataset
ds = load_dataset("Bingsu/zeroth-korean")