CoRal: Danish Conversational and Read-Aloud Speech with 100K+ Records

CoRal provides between 100,000 and 1,000,000 Danish audio recordings and transcriptions for Automatic Speech Recognition (ASR) tasks. Created by the CoRal-project and updated in early 2025, the collection includes both conversational and read-aloud speech samples across various dialects and age groups.

Use Cases

• Training ASR models using the audio and text transcription pairs
• Classifying speaker demographics using the age and gender labels
• Analyzing dialectal variations in Danish speech using the dialect and accent metadata

Strengths

• Scale of 100,000 to 1,000,000 records
• Dual modality featuring both conversational and read-aloud speech
• Inclusion of diverse Danish dialects and accents

Limitations

• Lack of detailed column documentation in the primary metadata
• Potential for label noise if transcriptions were not expert-verified

Provenance

Source

CoRal-project

Collection Method

Recorded speech (conversational and read-aloud)

Freshness

Last updated January 2025.

Geography

Denmark