Neuromorphic Continuous-Time State Space Pole Placement Adaptive Control

Abstract: A neuromorphic continuous-time state space pole assignment adaptive controller is proposed, which is particularlyappropriate for controlling a large-scale time-variant state-space model due to the parallely distributed nature ofneurocomputing. In our approach, Hopfield neural network is exploited to identify the parameters of a continuous-timestate-space model, and a dedicated recurrent neural network is designed to compute pole placement feedback control law inreal time. Thus the identification and the control computation are incorporated in the closed-loop, adaptive, real-timecontrol system. The merit of this approach is that the neural networks converge to their solutions very quickly andsimultaneously.     

基于REMOS的远距离语音识别模型补偿方法

封闭环境中远距离语音识别会受到混响效果的影响,从而降低语音识别率。混响建模(reverberation modeling for speech recognition,REMOS)是一种在模型域进行混响补偿的新方法,该方法在已知声源位置的情况下能有效提升远距离语音识别精度。但在实际应用中,往往难以预测声源的位置。利用最大后验概率的原理,基于对房间不同区域进行有区别补偿的思想,在按帧的隐马尔可夫模型 (hidden Markov model,HMM)补偿的基础上,提出一种在封闭环境中新的模型补偿方法。该方法利用K均值聚类K-means算法对房间冲击响应 (room impulse response,RIR)的优化集进行聚类,对所属相同类的混响模型进行合并处理,再把合并后的混响模型载入维特比算法中,对清晰语音的HMM模型进行按帧补偿。最后采用后验概率方法选择最佳补偿,使得模型域的混响补偿能最接近精确补偿。实验证明,该方法能进一步提升远距离语音识别的精度。