Dataset · Digital-Divide-Data
Khmer language speech corpus on Cambodian culture
Digital-Divide-Data/khmer-speech-dataset
Nearly 728 hours of Khmer audio recordings with manual transcription for speech recognition.
What's inside
The dataset contains 727.94 hours of audio-text pairs in Khmer (from 100k to 1M records with an average duration of 8 seconds). The recordings were manually read by 12 native speakers from Cambodia (5 women and 7 men) on given topics about local culture. Each file includes speaker metadata: gender, age group, and hometown.
What you can do with it
- Training and fine-tuning ASR models for Khmer speech recognition
- Classifying texts and audio by cultural themes and speaker characteristics
- Tasks
- Automatic speech recognition, Text classification
- Languages
- Khmer
- Size
- 100K–1M rows
- Downloads this month
- 9.7K
- Likes
- 38
- Access
- on request on Hugging Face
Models trained on it
- phonsobon/PsarAI-ASR-1BAutomatic speech recognition 96
- nphearum/PsarAI-ASR-1BAutomatic speech recognition 77
- KrorngAI/TrorYongASR2-smallAutomatic speech recognition 69
- KrorngAI/TrorYongASR2-tinyAutomatic speech recognition 25
- ric912/khmer-speech-text-tokenizer-4k 0
- Nukthean/krama-asr-khmerAutomatic speech recognition 0
- Nukthean/krama-asr-khmer-onnxAutomatic speech recognition 0
How to load it
from datasets import load_dataset
ds = load_dataset("Digital-Divide-Data/khmer-speech-dataset")