This dataset is designed for evaluating Uzbek speech-to-text (STT) models on real-world conversational speech data. The audio samples were collected from various open Telegram groups, capturing natural voice messages in diverse acoustic conditions and speaking styles.
