RoboCup@Home: Evaluation of voice recognition systems for domestic service robots and introducing Latino Dataset

摘要

Service robotics is a growing area with a huge impact on humans’ everyday life. RoboCup@Home competition has been pushing the limits of service robot research by presenting more challenging tests for the competitors every year. Fluid communication between service robot systems and humans is a must to improve robot performance. This challenging task has two main issues (i) the noisy environment, (ii) the speaker’s accents since most operators are non-native English speakers. In this paper, we present an evaluation of state-of-the-art ASR systems in the context of the RoboCup@Home competition. Also, we introduce the Latino Dataset, a novel dataset of RoboCup@Home-like commands from non-native Latin American speakers. Several experiments were conducted in order to simulate the RoboCup@Home competition environment. To this end, four datasets were chosen to cover the whole accents of English commands the robots are exposed (i) clean native English, (ii) native English with noise, (iii) non-native Asian English, and (iv) non-native Latin American English. The results show that PicoVoice and Kaldi would perform better in RoboCup@Home environments.