Khmer ASR Cultural: 134 Hours of Curated Cambodian Speech

134.6 hours of manually curated Khmer speech-text pairs focusing on Cambodian cultural topics, produced by DDD-Cambodia. The collection features recordings from 8 native speakers with an average duration of 8.54 seconds per utterance and includes detailed speaker metadata.

Use Cases

• Automatic Speech Recognition (ASR) training using Khmer speech-text pairs
• Speaker demographic classification using gender and age group metadata
• Cultural topic modeling using the topic and subtopic labels

Strengths

• 134.6 hours of curated audio
• 8 native speakers with 50/50 gender split
• Includes speaker metadata for age and origin city
• Manually verified transcriptions

Limitations

• Small speaker pool of only 8 individuals
• Geographic bias possible as speakers represent specific origin cities
• Domain-specific focus on culture

Provenance

Source

DDD-Cambodia

Collection Method

Manually recorded and curated by native speakers

Freshness

Last updated February 2026.

Geography

Cambodia