Central Thai and Dialect Speech Corpus with Parallel Sentences

700 hours of Central Thai speech and 40 hours each for three other Thai dialects form this corpus. The dataset, created by CMKL, includes parallel sentences across dialects to support speech and translation research. It was last updated in September 2024.

Use Cases

• Train automatic speech recognition models on 700 hours of Central Thai audio.
• Develop dialect adaptation systems using parallel sentences across the four Thai dialect recordings.
• Research cross-dialect machine translation leveraging aligned audio-text pairs.
• Benchmark speech model performance on Thai dialects using the provided data splits.

Strengths

• 700 hours of Central Thai audio provides a substantial training resource.
• Includes parallel sentences across four dialects for controlled comparison.
• Officially split data partitions support reproducible evaluation.

Limitations

• Dialect data is smaller at 40 hours per dialect, limiting model training scope.
• Specific details on speaker demographics, recording conditions, and label quality are not provided.

Provenance

Source

CMKL, with Central Thai data collected via Wang Data Market.

Collection Method

null

Time Range

null

Freshness

Last updated September 2024.

Geography

Thailand, focusing on Central and three other dialect regions.