Khmer YouTube Voice Dataset with 1,000+ Hours of Diarized Speech

Khmer Yt Voice Dataset contains 3,945 YouTube videos totaling 1,033.4 hours of Khmer speech audio. The dataset includes diarization metadata for 16,021 speaker turns with transcripts. It was created by author manhp and last updated on 2026-04-19.

Use Cases

• Train automatic speech recognition models based on the provided audio transcripts.
• Develop speaker diarization systems based on the segmented speaker turn metadata.
• Build Khmer text-to-speech models based on the extensive collection of native speaker audio.
• Conduct linguistic analysis of Khmer speech patterns based on the YouTube-sourced content.

Strengths

• Contains 1,033.4 hours of audio, providing substantial material for model training.
• Includes diarization metadata for 16,021 speaker turns, enabling speaker-specific analysis.
• Comprises 3,945 source videos, suggesting diversity in content and speakers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic or content bias inherent to its source, YouTube.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

YouTube videos.

Collection Method

Audio extracted and diarized from Khmer-language YouTube videos.

Freshness

Last updated 2026-04-19 01:25:31; freshness should be verified.

Geography

Likely Cambodia or Khmer-speaking regions, given the language focus.