Wake word and speech recognition application on edge device: a case of improving the electric wheelchair / Low Jian He ... [et al.]

• Main Authors: Low, Jian He, Fadhlullah, Solahuddin Yusuf, Kamarulazizi, Khadijah, Abdullah, Samihah, Abdul Hamid, Shabinar
• Format: Article
• Language: English
• Published: UiTM Press 2024
• Subjects: Applications of electronics
• Online Access: https://ir.uitm.edu.my/id/eprint/94745/1/94745.pdf; https://ir.uitm.edu.my/id/eprint/94745/

Edge devices play an ever-increasing role as to reduce latency, improve efficiency and adapt simplicity. However, their lower processing capabilities compared to traditional setup means that their usage are also limited. This research explores the possibility of implementing both wake word and speech recognition on an edge device. The proposed wake word system, which is for voice activation, is developed based on the LSTM Neural Network model. The model is trained, modelled, and evaluated to respond to the wake word of “Hey SellTron”. As for the speech recognition (voice command) system, Google Speech API was selected to recognize standard directional commands (left, right, forwards, backwards), as well as the new conceptual locational commands (“go to kitchen”, “go to bedroom”, etc.). To prove the feasibility of the developed system, these two features were integrated into an edge device on a mobile platform; representing a conceptual mini wheelchair ‘prototype’ due to project constraints. Evaluations showed that the prototype was able to activate and respond to voice commands correctly with over 80% accuracy in low ambient noise (<50dB).