Name entity recognition for Air Traffic Control transcripts using deep learning based approach

Voice communication is the main way that pilots and Air Traffic Controllers (ATCos) exchange information. Recently, many studies have looked into the use of Automatic Speech Recognition (ASR) technology to automatically extract important information which has the potential to lessen the workload of ATCos. ATC Transcription provides textual information for radio channel dialogue. Due to regional differences, this textual data does not strictly adhere to phraseology and contains transcription errors making it challenging to identify entities or meaningfully annotate the data. The research mainly focuses on NER on ATC transcripts provided by Honeywell Technology Solutions Pvt. Ltd. using different deep learning-based classifiers like GRU, LSTM, Bi-LSTM, Bi-GRU, BERT, Bi-LSTM & CRF. BIO tagging is applied on the dataset using regular expression before providing to deep learning models. All the models were implemented and each model was analyzed by entity level classification and obtained good results but BERT and Bi-LSTM and CRF performed well when compared to other models with f1-score and precision of approximately 98% and more.