With the rise of Artificial Intelligence and Internet of Things, more and more applications are designed to bring benefit for our daily life. Voice is an easily generated input method by human. The process of converting voice into text is speech recognition. Online tools could provide high accuracy for speech recognition. In some applications, e.g., factory, Internet is forbidden and the number of specific voice commands are limited. It becomes necessary to be able to perform speech recognition in real-time on the embedded devices without having the need to send to the remote servers. In this paper, we enhance a speech-to-text framework (PocketSphinx) on embedded devices, i.e. Raspberry PI. The enhanced PocketSphinx can execute speech-to-text on embedded devices without Internet. Moreover, it can also recognize hybrid language (English and Chinese) to control a IoT or ROS-based device.