TVSpeech is a Thai speech recognition benchmark dataset specifically designed as a Robustness Track for evaluating ASR models on real-world, in-the-wild Thai audio. The dataset consists of 570 utterances (3.75 hours) curated from diverse public media channels on YouTube under the Creative Commons Attribution (CC-BY) license, representing challenging acoustic and semantic complexity found in natural speech.
Language: Thai… See the full description on the dataset page:
https://huggingface.co/datasets/typhoon-ai/TVSpeech.
