Enhancing Keyword Spotting in Noisy Environments: A Deep Learning Approach with MFCC and Bidirectional LSTM Networks

Keyword spotting (KWS) plays a pivotal role in voice -assisted technologies, enabling prompt system activation upon detecting specific keywords in a user's speech. However, robust KWS in noisy environments remains a challenge. This research work presents a novel deep learning -based approach for keyword spotting, leveraging mel-frequency cepstral coefficients (MFCC) and Bidirectional Long Short -Term Memory (BiLSTM) networks. The proposed system extracts MFCC features from audio samples and employs a BiLSTM network to capture long-term dependencies in time sequences, analyzing both forward and backward contexts. The network is trained on the Google Speech Commands Dataset, and data augmentation techniques are utilized to enhance its performance in noisy conditions. The effectiveness of the trained model is evaluated, demonstrating its ability to accurately spot keywords amidst background noise. The research work offers valuable insights into improving voice-assist technologies and contributes to advancing keyword spotting techniques in challenging real-world scenarios.