Bangla999-ECD: A Simulated Bengali Emergency Call Transcript Dataset

The dataset contains approximately 24,000 text-only call transcripts generated by three native Bengali speakers, covering four categories: Non-emergency, Police, Fire, and Medical emergencies. Each record includes a Bangla transcript of a simulated 999 emergency call, the cleaned version after basic preprocessing, an optional contextual description, a binary emergency indicator(is_emergency), and a multi-class emergency type label.

All samples are synthetic and do not contain real emergency calls, personal identifiers, or sensitive information. The dataset was curated using dataset-level preprocessing steps, including Unicode normalization, punctuation and stopword removal, duplicate elimination using MD5 hashing, and removal of incomplete records. The cleaned dataset is provided in CSV format to facilitate reuse in natural language processing, emergency response research, and Bangla text classification tasks.