Zero-crossing-based linear prediction for speech recognition

The satisfactory estimation of speech autocorrelation by means of generalised zero-crossings indicates that they can be used for efficient feature extraction in speech recognition. In addition, high consistency between the Itakura-Saito distances, calculated before and after clipping, allowed for only a mode-rate degradation of the related recognition performance, which was compensated by including the excitation distortion into the distance measure.<>     

Robust backward adaptive formant prediction for speech coder

To improve the error performance of speech coders, an adaptation method for the backward adapted formant predictor is proposed. The filtered residual signal is used instead of the reconstructed output signal as the input to an adaptation of the formant predictor. The performance of the filtered-residual driven adaption method in the noise free channel is as good as that of conventional output driven adaptation. Moreover, the new adaptation method maintains the same robustness to channel errors as residual driven adaptation     

Neural networks for statistical recognition of continuous speech

In recent years there has been a significant body of work, both theoretical and experimental, that has established the viability of artificial neural networks (ANN's) as a useful technology for speech recognition. It has been shown that neural networks can be used to augment speech recognizers whose underlying structure is essentially that of hidden Markov models (HMM's). In particular, we have demonstrated that fairly simple layered structures, which we lately have termed big dumb neural networks (BDNN's), can be discriminatively trained to estimate emission probabilities for an HMM. Recently simple speech recognition systems (using context-independent phone models) based on this approach have been proved on controlled tests, to be both effective in terms of accuracy (i.e., comparable or better than equivalent state-of-the-art systems) and efficient in terms of CPU and memory run-time requirements. Research is continuing on extending these results to somewhat more complex systems. In this paper, we first give a brief overview of automatic speech recognition (ASR) and statistical pattern recognition in general. We also include a very brief review of HMM's, and then describe the use of ANN's as statistical estimators. We then review the basic principles of our hybrid HMM/ANN approach and describe some experiments. We discuss some current research topics, including new theoretical developments in training ANN's to maximize the posterior probabilities of the correct models for speech utterances. We also discuss some issues of system resources required for training and recognition. Finally, we conclude with some perspectives about fundamental limitations in the current technology and some speculations about where we can go from here     

Neural network architectures for vector prediction

A vector predictor is an integral part of a predictive vector quantization coding scheme. The conventional techniques for designing a nonlinear predictor are extremely complex and suboptimal due to the absence of a suitable model for the source data. We investigated several neural network architectures that can be used to implement a nonlinear vector predictor, including the multilayer perceptron, the functional link network and the radial basis function network. We also evaluated and compared the performance of these neural network predictors with that of a linear vector predictor. Our experimental results show that a neural network predictor can predict the blocks containing edges with a higher accuracy than a linear predictor. However, the performance of a neural network predictor is comparable to that of a linear predictor for predicting the stationary and shade blocks     

Neural networks for vector quantization of speech and images

Using neural networks for vector quantization (VQ) is described. The authors show how a collection of neural units can be used efficiently for VQ encoding, with the units performing the bulk of the computation in parallel, and describe two unsupervised neural network learning algorithms for training the vector quantizer. A powerful feature of the new training algorithms is that the VQ codewords are determined in an adaptive manner, compared to the popular LBG training algorithm, which requires that all the training data be processed in a batch mode. The neural network approach allows for the possibility of training the vector quantizer online, thus adapting to the changing statistics of the input data. The authors compare the neural network VQ algorithms to the LBG algorithm for encoding a large database of speech signals and for encoding images     

Neural networks for modeling nonlinear memoryless communicationchannels

This paper presents a neural network approach for modeling nonlinear memoryless communication channels. In particular, the paper studies the approximation of the nonlinear characteristics of traveling-wave tube (TWT) amplifiers used in satellite communications. The modeling is based upon multilayer neural networks, trained by the odd and even backpropagation (BP) algorithms. Simulation results demonstrate that neural network models fit the experimental data better than classical analytical TWT models,     

Helium speech processor using linear prediction

Currently available helium speech processors provide a linear frequency scaling but, at the greater depths encountered in recent years, a nonlinear shift is desirable. Here, we suggest a method based on linear prediction, which will produce such a shift, while retaining a reasonably low computational cost.     

France's grass roots data net

The authors describe how France introduced the Minitel information highway network in 1982, and now 80 percent of the French nation's businesses and 20 percent of its citizens use it. Network addresses are equivalent to 800 freephone numbers in the United States, but Minitel still manages to run at a profit     

Neural network architecture for solving nonlinear equation systems

A general neural network architecture is derived which can achieve ultrahigh-speed computation in solving large nonlinear equation systems. The computing time required for a solution is independent of the dimension of the equation system which is solved. The approach to solving equation systems by the method reported is very different from that of implementing in software a numerical analysis procedure and solution algorithm.<>     

Adaptive prediction in speech differential encoding systems

The design of speech coders that produce high-quality highly intelligible speech at 6 to 16 kb/s while retaining robustness to background and transmission impairments is an area of current research interest. Differential encoding structures employing adaptive quantization and adaptive prediction constitute one of the most promising approaches to achieving these design objectives. This paper focuses on the design and analysis of adaptive predictors for differential encoders. Several differential encoding systems, including adaptive predictive coding, differential pulse-code modulation, noise feedback coding, direct feedback coding, and prediction error coding, are described and related. Adaptive quantizers are briefly discussed and quantitative and qualitative indicators of speech coder performance are defined. The channel model, the speech model, and the research problem statements used in the design of differential encoders and adaptive predictors are presented. The nomenclature and theory of forward and backward adaptive prediction are developed, and several new backward adaptive algorithms based on various assumptions are presented. A detailed survey of theoretical and simulation results on adaptive prediction for speech differential encoders is given, and the effects of background and transmission impairments on these systems are discussed, Finally, the impact of adaptive predictors on rate distortion theory motivated coders is indicated. Numerous areas for future research are highlighted.     

Waveform coding for low-power digital filtering of speech data

This paper evaluates waveform coding techniques known from low bit-rate communication for their usefulness in low-power digital FIR filtering of speech signals. The encodings considered include linear PCM, PCM with adaptive and logarithmic quantization, and differential PCM, combined with two's-complement and sign-magnitude number representation. Selected implementation aspects for each alternative are discussed. Experimental results are presented to quantify potential power savings subject to statistical signal properties and operating conditions. Guidelines for the choice of encoding in application-specific digital signal processing of speech data are provided.     

A speech coder for PC multimedia net‐to‐net communication

The paper presents a speech coding algorithm for operation at 11025 samples/s. The coder provides improved speech quality and compatibility with the MS‐Windows multimedia environment. The coding algorithm has been developed by adapting the ITU G729 and enhancing it with some recent developments in the medium band coding. The coder operates over a band of frequencies ranging from 20 to 5400 Hz at a bit rate of 8.9 kbit/s. Application of this coder includes intranet VoIP, voice chatting, multimedia communications, and voice archiving. Copyright © 2001 John Wiley & Sons, Ltd.     

Vector quantisation for minimum residual energy in pitch prediction of speech

A method for vector quantisation of pitch predictor coefficients according to a minimum residual energy criterion is proposed and compared to vector quantisation using the traditional minimum squared error between coefficients. Squared error quantisation is found to be adequate for 1-tap prediction, but for 3-tap prediction the residual energy method performs consistently better. The predictor code-books are also found to give robust performance outside the training sequence.     

Genetic algorithm for optimizing the nonlinear time alignment ofautomatic speech recognition systems

Dynamic time warping (DTW) is a nonlinear time-alignment technique for automatic speech recognition (ASR) systems. It had been widely used in many commercial and industrial products, ranging from electronic dailies/dictionaries to wireless voice digit dialers. DTW has the advantages of fast training and searching times, which makes it more popular than other available ASR techniques. However, there exist some limitations to DTW, such as the stringent rule on slope weighting, the nontrivial computation of the K-best paths, and the significant increase in computational time when the endpoint constraint is relaxed or the variations of the length of pattern increased. In this paper, a stochastic method called the genetic algorithm (GA), which is used to solve the nonlinear time alignment problem, is presented. Experimental results show that the GA has a better performance than the DTW. In addition, two derivatives of GA: the hybrid GA and the parallel GA are also presented     

A speech/voiceband data discriminator

The problem of distinguishing speech from voiceband data is treated. A discrimination function based on the sign of the autocorrelation at lag two of the incoming signal and the second-order moment of the complex low-pass signal is presented. The algorithm has been applied to many types of voiceband data signals and, within a window of 32 ms, correctly classifies them. For the limited number of speech signals available, the amount of misclassification of speech as voiceband data was observed to be about 1%     

非线性系统神经网络L2增益控制

针对很难求解Ⅲ偏微分不等式解析解的问题,采取一种新的思想,给出了利用神经网络构造HJ不等式解的一种方法,并利用遗传算法进行神经网络权值的优化,使得Lyapunov函数满足HJ不等式,避免了求解HJ偏微分不等式。并在此基础上,给出基于HJ不等式的仿射非线性系统神经网络L2增益抗干扰控制器设计方法,以神经网络的形式给出厶增益控制器的一般结构。仿真结果表明,提出的控制器设计方法是可行的,实现了闭环系统为从外界干扰到系统输出是有限增益L2稳定的。     

非线性系统神经网络L_2增益控制

针对很难求解HJ偏微分不等式解析解的问题,采取一种新的思想,给出了利用神经网络构造HJ不等式解的一种方法,并利用遗传算法进行神经网络权值的优化,使得Lyapunov函数满足HJ不等式,避免了求解HJ偏微分不等式.并在此基础上,给出基于HJ不等式的仿射非线性系统神经网络L2增益抗干扰控制器设计方法,以神经网络的形式给出L2增益控制器的一般结构.仿真结果表明,提出的控制器设计方法是可行的,实现了闭环系统为从外界干扰到系统输出是有限增益L2稳定的.     

Attack path prediction method based on causal knowledge net

The existing attack path prediction methods can not accurately reflect the variation of the following attack path caused by the capability of the attacker.Accordingly an attack path prediction method based on causal knowledge net was presented.The proposed method detected the current attack actions by mapping the alarm sets to the causal knowledge net.By analyzing the attack actions,the capability grade of the attacker was inferred,according to which adjust the probability knowledge distribution dynamically.With the improved Dijkstra algorithm,the most possible attack path was computed.The experiments results indicate that the proposed method is suitable for a real network confrontation environment.Besides,the method can enhance the accuracy of attack path prediction.     

Sensor data fusion using perception net for a precise assembly task

A sensor fusion method is presented for a peg-in-hole insertion task. Three kinds of sensor are fused for task execution. The vision and proximity sensors are mainly used for gross motion control, and the force/torque sensor is used for fine motion control of the robot. Covariance analysis is conducted for each sensor discussed in this paper, and the perception net is introduced to improve the usefulness of each sensor by optimally weighting its output.