Published Paper Details:

Text-Based Spoken Term Detection (STD) is the task of detecting and temporally localizing user-defined textual query terms directly from large collections of unstructured speech data. Unlike Automatic Speech Recognition (ASR), which focuses on generating complete transcriptions, STD concentrates on accurately identifying specific keywords with minimal false alarms and missed detections. Over the last two decades, STD has evolved from template-based methods and HMM-GMM frameworks to modern deep learning and transformer-based architectures. This paper presents a comprehensive survey of text-based STD systems covering literature evolution, system architecture, feature extraction, keyword modeling, ASR-based and acoustic-based techniques, deep learning advancements, similarity matching strategies, standard evaluation metrics, benchmarking tools, key challenges, and real-world applications. In addition, a complete implementation framework using the QUESST-2014 dataset with the Terrier Information Retrieval platform and the Hiemstra Language Model is presented, with speech transcription performed using the HappyScribe platform.