Khmer ASR Cultural Dataset
Digital Divide DataKhmer ASR Cultural Dataset is a speech corpus of about 728 hours of Khmer read by native speakers, paired with transcripts, for training speech recognition and related models. Twelve speakers from Cambodia (five women, seven men) recorded sentences written around 61 cultural topics and 1,749 subtopics, averaging eight seconds per clip, with speaker gender, age group and origin city attached. Digital Divide Data built and published it.
Openness
3 high confidence- license
- cc-by-sa-4.0
- access
- manual (Hugging Face gate with manual approval that shares the requester's username and email; the smaller V2 release on Mozilla Data Collective needs an account and accepted terms)
- dataset_card
- present (speakers, topics, format and sample rows on the Hub page; the raw README returns 401 behind the gate)
The audio and transcripts carry a share-alike license that asks users to credit Digital Divide Data, and the card describes speakers, topics and file layout. The Hugging Face copy now opens only after the publisher approves a request that shares the user's username and email, and a smaller, earlier release on Mozilla Data Collective needs an account and accepted terms.
- https://huggingface.co/api/datasets/Digital-Divide-Data/khmer-speech-dataset recorded 2026-11-12
API JSON: "gated":"manual"; tag license:cc-by-sa-4.0.
- https://huggingface.co/datasets/Digital-Divide-Data/khmer-speech-dataset recorded 2026-11-12
Gate box: "You need to agree to share your Hugging Face information to access this dataset"; "you have to accept the conditions to access its files and content"; page data "gated":"manual" with "gatedProfileSharing":{"username":true,"email":true}. Card: "license is Creative Commons Attribution Share Alike 4.0 International (CC-BY-SA-4.0). Please attribute"; "Speaker metadata (gender, age group, and origin city) is provided."; "with a total of 61 topics and 1,749 subtopics".
- https://huggingface.co/datasets/Digital-Divide-Data/khmer-speech-dataset/raw/main/README.md recorded 2026-11-12
HTTP 401, "Access to dataset Digital-Divide-Data/khmer-speech-dataset is restricted. You must have access to it and be authenticated to access it." The card file sits behind the gate.
- https://mozilladatacollective.com/datasets/cml9h5vgc01bxmn075sjeftek recorded 2026-11-12
"Khmer ASR Cultural Dataset (V2)"; "106.53 hours manually curated speech-text pairs"; "4 females, 4 males"; "Creative Commons Attribution Share Alike 4.0 International (CC-BY-SA-4.0)"; "You agree not to attempt to determine the identity of speakers in this dataset."; "You will only be able to download datasets after accepting terms".
Adoption
2 high confidenceAdoption is Hugging Face downloads of the repository. Copies taken from the Mozilla Data Collective listing are not included.
- https://huggingface.co/api/datasets/Digital-Divide-Data/khmer-speech-dataset recorded 2026-09-24
3787 downloads in the trailing 30 days for Digital-Divide-Data/khmer-speech-dataset
Capability
3 medium confidenceThe corpus dwarfs OpenSLR's earlier Khmer set of under four hours, pairing prompted readings with curated transcripts, and a community Qwen recognizer names it as its training data. Its seven hundred hours from twelve speakers are judged against what Khmer has, an official language with over a million speakers and a small written web, and on that scale it sits one step below the larger Kazakh Speech Corpus 2.
- https://huggingface.co/datasets/Digital-Divide-Data/khmer-speech-dataset recorded 2026-11-11
"727.94 hours of manually curated speech-text pairs by native speakers"; "there was only one Khmer speech-text dataset:"; "with 3.97 hours of speech-text pairs (male only)"; the "Models trained or fine-tuned on" panel lists "seanghay/Qwen3-ASR-0.6B-Khmer".
- https://huggingface.co/seanghay/Qwen3-ASR-0.6B-Khmer/raw/main/README.md recorded 2026-11-11
"A Khmer (ខ្មែរ) automatic speech recognition model, fine-tuned from"; "~700 hours of Khmer speech from the"; trained on DDD-Cambodia/khmer-speech-dataset.
Verified 2026-11-11