Presenter: Dr. Nguyen Tien Hung

Title: Voice-controlled robot based on the AI and TensorFlow Lite

Content: When machines or robots understand human voices, people can communicate with machines or control them more easily. In a real-world, speech recognition would be difficult because each person’s voice is different. Even with the same word from the same person speaking but in different environments, the received voice will be different due to the influence of environmental noises. Therefore, in order to identify words spoken by different people in different environments, it is necessary to collect a large amount of speech data (data set). This dataset is then analyzed to find the features of the words (feature extraction). This feature dataset is then used for identification. After that, a newly recorded speech sample will be used as an input signal to the speech recognition system. The speech recognizer implements an algorithm that can make a speech prediction based on the input speech pattern. In this report, we will show how to use a free Google speech dataset to train a machine learning model and thereby serve for word recognition later. The tool used for training is the TensorFlow platform. Then, the TensorFlow Lite engine is applied to convert the obtained model into a minimalistic form that can be implemented on embedded systems. Finally, the research results are demonstrated by controlling the movement of a Robot in the forward, backward, left, and right directions.