Simultaneous perturbation stochastic approximation （SPSA） belongs to the class of gradient-free optimization methods that extract gradient information from successive objective function evaluation. This paper descri...Simultaneous perturbation stochastic approximation （SPSA） belongs to the class of gradient-free optimization methods that extract gradient information from successive objective function evaluation. This paper describes an improved SPSA algorithm, which entails fuzzy adaptive gain sequences, gradient smoothing, and a step rejection procedure to enhance convergence and stability. The proposed fuzzy adaptive simultaneous perturbation approximation （FASPA） algorithm is particularly well suited to problems involving a large number of parameters such as those encountered in nonlinear system identification using neural networks （NNs）. Accordingly, a multilayer perceptron （MLP） network with popular training algorithms was used to predicate the system response. We found that an MLP trained by FASPSA had the desired accuracy that was comparable to results obtained by traditional system identification algorithms. Simulation results for typical nonlinear systems demonstrate that the proposed NN architecture trained with FASPSA yields improved system identification as measured by reduced time of convergence and a smaller identification error.展开更多
The first step of missing feature methods in text-independent speaker identification is to identify highly corrupted spectrographic representation of speech as missing feature. Most mask estimation techniques rely on ...The first step of missing feature methods in text-independent speaker identification is to identify highly corrupted spectrographic representation of speech as missing feature. Most mask estimation techniques rely on explicit estimation of the characteristics of the corrupting noise and usually fail to work with inaccurate estimation of noise. We present a mask estimation technique that uses neural networks to determine the reliability of spectrographic elements. Without any prior knowledge of the noise or prior probability of speech, this method exploits only the characteristics of the speech signal. Experiments were performed on speech corrupted by stationary F16 noise and non-stationary Babble noise from 5dB to 20 dB separately, using cluster based reconstruction missing feature method. The result performs better recognition accuracy than conventional spectral subtraction mask estimation methods.展开更多
文摘Simultaneous perturbation stochastic approximation （SPSA） belongs to the class of gradient-free optimization methods that extract gradient information from successive objective function evaluation. This paper describes an improved SPSA algorithm, which entails fuzzy adaptive gain sequences, gradient smoothing, and a step rejection procedure to enhance convergence and stability. The proposed fuzzy adaptive simultaneous perturbation approximation （FASPA） algorithm is particularly well suited to problems involving a large number of parameters such as those encountered in nonlinear system identification using neural networks （NNs）. Accordingly, a multilayer perceptron （MLP） network with popular training algorithms was used to predicate the system response. We found that an MLP trained by FASPSA had the desired accuracy that was comparable to results obtained by traditional system identification algorithms. Simulation results for typical nonlinear systems demonstrate that the proposed NN architecture trained with FASPSA yields improved system identification as measured by reduced time of convergence and a smaller identification error.
文摘The first step of missing feature methods in text-independent speaker identification is to identify highly corrupted spectrographic representation of speech as missing feature. Most mask estimation techniques rely on explicit estimation of the characteristics of the corrupting noise and usually fail to work with inaccurate estimation of noise. We present a mask estimation technique that uses neural networks to determine the reliability of spectrographic elements. Without any prior knowledge of the noise or prior probability of speech, this method exploits only the characteristics of the speech signal. Experiments were performed on speech corrupted by stationary F16 noise and non-stationary Babble noise from 5dB to 20 dB separately, using cluster based reconstruction missing feature method. The result performs better recognition accuracy than conventional spectral subtraction mask estimation methods.
