Enhancing Keyword Spotting Performance Using Hexadecimal Encoding for Edge Devices

Keyword spotting is used in a wide range of applications, including home automation, voice assistants, audio processing, and more. Accuracy is a key factor in these applications, and the same has been the motive in this paper. Memory and processing complexity also play a pivotal role in such deployments. Currently, most keyword spotting methodologies rely on direct spectrogram-based feature extraction from raw audio input to train the CNN model. To improve performance metrics, we have introduced Hexadecimal encoding-based feature extraction. This paper examines several conventional methods of encoding and how our method outperforms various metrics that can be used in further advancements in the audio processing of edge devices.