---
dataset_info:
  features:
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: transcription
    dtype: string
  splits:
  - name: train
    num_bytes: 3000520612.957125
    num_examples: 4569
  - name: validation
    num_bytes: 295462706.3865031
    num_examples: 324
  - name: test
    num_bytes: 720704006.3774319
    num_examples: 765
  download_size: 3638331196
  dataset_size: 4016687325.7210603
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

- Combined [openslr/openslr-SLR42](https://huggingface.co/datasets/openslr/openslr) and [google/fleurs-km-kh](https://huggingface.co/datasets/google/fleurs).
- Each audio is shorter than or equal to 30 seconds.
- Sampling rate is 16,000
- Each transcription is already normalized by `tha.normalize.processor` (`pip install tha`)