To deploy trained speech with DNN-LSTM framework for controlling a smart wheeled-robot in limited learning circumstance

The assessment to a trained speech controller with deep neural network long-short term memory (DNN-LSTM) framework adopted as the commander to control a smart wheeled-robot is implemented in the article. Accordingly, the deployment is implemented in recognition to control remotely a smart wheeled-robot which has been completed previously by other project of authors’ research group. Based on the machine learning skill a framework established with the DNN-LSTM model is embedded into the smart wheeled-robot prototype. Apart from, the control commands are designed over limited learning circumstance where constrained single-track (ST) and double-track (DT) speeches, and only are including 4 Chinese speech commands, “forward” (Chinese”前進”), “backward” (Chinese”後退”), “turn left” (Chinese”左轉”), and “turn right” (Chinese”右轉”). Though, there are just 4 simple speeches collected for data training, the investigation to the accurate ratio is deployed in 3 separated persons training work and each with 1000 to 5000 training times. There are just three parameters (this why “Limited Learning Circumstance” is referred as the article name) considered as the dominators for the performance evaluation of the speech controlled wheeled-robot. The results from the testing cases clearly show that the set with DT has the higher accurate comparison with the set of ST. The best outcomes form the performance of testing and validation happens at the case of DT channel, hereafter, the accurate and loss rate are obtained as 0.673 and 0.018 with 50% dropout, respectively. However, the ratio of dropout has been discovered definitely to dominate the accurate and loss rate when it is deployed during the process of training step. Eventually, the trained and developed model of speech command sets are uploaded into a micro-controller after accuracy analyzed, and embedded into the smart wheeled-robot plays as remotely pilot scheme.