Vaja-Thai: Combined Thai Speech Dataset for TTS Research, 554.6 Hours

Vaja-Thai is a unified Thai speech dataset containing 289,916 audio samples totaling 554.6 hours for Text-to-Speech research. The dataset was created by dubbing-ai and last updated in April 2026. All audio is resampled to 24 kHz WAV format and combines multiple quality-filtered sources.

Use Cases

• Training Thai speech synthesis models based on the 554.6 hours of audio.
• Fine-tuning TTS systems based on the unified, quality-filtered dataset.
• Benchmarking TTS model performance on a standardized Thai speech corpus.
• Researching prosody and pronunciation in Thai language based on the high-volume audio samples.

Strengths

• Contains 289,916 audio samples, providing substantial volume for model training.
• Totals 554.6 hours of Thai speech, offering extensive coverage for TTS tasks.
• Audio is standardized to 24 kHz WAV format, ensuring consistent technical specifications.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The license for the combined dataset is unknown, which may restrict usage.
• Data may reflect source bias inherent to the original contributing datasets.

Provenance

Source

dubbing-ai on Hugging Face, combining multiple sources including tsync2.

Collection Method

Unified and quality-filtered combination of multiple Thai speech sources.

Time Range

null

Freshness

Last updated 2026-04-01 19:14:14; freshness should be verified.

Geography

null