TVSpeech: Thai Video Speech Recognition Benchmark with 570 Utterances

TVSpeech is a Thai speech recognition benchmark dataset designed as a Robustness Track for evaluating ASR models. It contains 570 utterances totaling 3.75 hours of audio curated from diverse public YouTube channels under the Creative Commons Attribution license. The dataset was created by typhoon-ai and last updated on 2026-01-21.

Use Cases

• Benchmarking ASR model robustness based on real-world, in-the-wild Thai audio described in the dataset.
• Training speech recognition models on challenging acoustic and semantic complexity found in natural Thai speech.
• Evaluating model performance on diverse public media sources as mentioned in the dataset description.

Strengths

• Contains 570 utterances, providing a specific volume of data for evaluation.
• Totals 3.75 hours of audio, offering a defined duration of speech material.
• Curated from diverse public media channels on YouTube, suggesting varied acoustic conditions.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2026-01-21 03:32:22; freshness should be verified.

Provenance

Source

typhoon-ai via Hugging Face

Collection Method

Curated from diverse public media channels on YouTube under the Creative Commons Attribution (CC-BY) license.

Freshness

2026-01-21 03:32:22

Geography

Thailand (implied by language)