An AI-powered English pronunciation training assistant using NVIDIA Riva and pygame

Abstract

Developing correct English pronunciation requires practice in order to build skill. The objective of this research is to apply artificial intelligence technology from NVIDIA Riva, which provides speech recognition and speech synthesis, as a tool for human–computer interaction. A large language model (LLM) is used to function as an intelligent agent that communicates through spoken dialogue, and pygame is employed to create a graphical interface in the form of an animated robot face. ROS2 serves as the middleware connecting the various system components. The system’s goal is to support English pronunciation training and to foster positive attitudes toward English learning.

The system is designed to run specifically on the Jetson Orin NX 16GB, a small, energy-efficient yet high-performance computer. This enables the system to operate offline, without relying on a network, which improves processing speed and enhances security—unlike typical cloud-based systems that often suffer from higher latency and lower security.

               Testing results show that the system performs well, with low response times and high processing accuracy from NVIDIA Riva. In conclusion, the techniques presented in this research are highly suitable for developing innovative conversational intelligent agents, offering significant potential benefits across many future applications.