基于HMM和RNN的无人机语音控制方案与仿真研究

为简化无人机操作,避免误操作,设计了一套基于隐马尔可夫模型(Hidden Markov Model,HMM)和循环神经网络(Recurrent Neural Networks,RNN)的无人机语音控制方案。该方案采用HMM识别无人机语音指令;同时采用RNN对多套无人机操作指令串进行训练,并对当前时刻指令进行预测,通过计算二者的相关性判断是否执行。仿真结果表明,该方案对HMM识别错误指令的辨别率达到61.90%,使整体错误率降至1.43%,表明该方案具有较为优异的性能。

Speech Control Scheme Design and Simulation for UAV Based on HMM and RNN

In order to simplify the operation and avoid the misoperation of UAVs, Based on Hidden Markov Model and Recurrent Neural Networks, a speech control scheme for UAVs is designed. In this scheme, HMM is used to train and recognize the speech command samples of UAVs. HMM is used to pick out the error commands, RNN is used to train the sets of UAVs commands, and the next command based on the training result is predicted. It is determined whether to execute or not by calculating the correlation between commands recognized by HMM and predicted by RNN. The simulation results show the recognition rate of wrong command is as high as 61.90%, and the overall error rate is down to 1.43%. All show the excellent performance of this scheme.