PuristanLabs1/Urdu-Turn-Detection-10k

一个高质量的乌尔都语句子数据集，包含10,000个句子，用于对话转向检测（End-of-Turn）。该数据集旨在帮助对话AI系统判断用户是否已完成发言（Complete）或正在暂停/犹豫（Incomplete）。数据集包含真实对话和合成数据，语言为纯乌尔都语（Nastaliq/Arabic Script），无英语或罗马字符污染。标签为1（Complete）和0（Incomplete），平衡性约为50%。适用于训练低延迟分类模型（如BERT、DistilBERT），用于实时语音机器人和代理。

A high-quality dataset of 10,000 Urdu sentences labeled for Turn Detection (End-of-Turn). This dataset is designed to help conversational AI systems determine if a user has finished speaking (Complete) or is pausing/trailing off (Incomplete). The dataset is a combination of real-world validation and high-quality synthetic generation, with pure Urdu script (Nastaliq/Arabic Script) and no English/Roman script contamination. Labels are 1 (Complete) and 0 (Incomplete), with a balance of ~50%. Optimal for training low-latency Classification models (BERT, DistilBERT) for realtime voice bots & Agents.