Bangla999-ECD: A Simulated Bengali Emergency Call Transcript Dataset

Description

Bangla999-ECD is a simulated Bangla (Bengali) emergency call transcript dataset created to support research in emergency calls and non emergency(or prank) calls classification for Bengali language. 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.

Steps to reproduce

All data processing steps used to generate the cleaned version of the Bangla999-ECD dataset are documented and reproducible. The repository includes the original raw dataset, a cleaned dataset, and a preprocessing script describing dataset-level cleaning operations such as text normalization, duplicate removal, and handling of missing values. The dataset can be reproduced by executing the provided preprocessing script on the raw data using Python (version ≥3.8) with standard scientific libraries (pandas, numpy). A recommended stratified train/validation/test split (80/10/10) is documented to support reproducible experimental evaluation.

Institutions

• Premier University

  Chittagong, Chittagong

Categories

Licence