Applying a Deep Learning Technique for Speech Recognition in Robotics

摘要

In times of crisis, such as a pandemic, technological breakthroughs are more likely to occur. And this is what is happening in the world right now, where artificial intelligence and robotics are experiencing unprecedented progress encouraged by the countries. In this paper, we describe a system that gives a mobile robot the ability to perform automatic speech recognition (ASR) with multi speakers. So, it is necessary to conduct a thorough investigation into vocal recognition systems additionally deep learning techniques applicable to this field. In contrast, automatic speech recognition faces two significant challenges: selecting relevant signal parameters that denote the speech signal and identifying a suitable decoding process that provides reliable recognition. In the context of this work, we investigate the use of a deep learning technique for speech recognition that can recognize ten voices (off, on, left, right, down, up, go, backward, forward, and stop). The voice command allows the mobile robot to be guided and controlled using the Raspberry Pi model B card and five DC motor drives. A real-time system was implemented and configured using an offline Wi-Fi network between software and hardware components. The whole system has been evaluated based on an English speech corpus trained and validated by native Arabic speakers for single words to assess real-time performance. As a result of this decreasing accuracy score, we have created a new dataset containing records of members of our research laboratory. The results showed an accuracy of approximately 89.27 % for the prediction of all ten voice commands.