By Topic

Acoustics, Speech, and Signal Processing, 1996. ICASSP-96. Conference Proceedings., 1996 IEEE International Conference on

Date 9-9 May 1996

Go

Filter Results

Displaying Results 1 - 25 of 139
  • Authors index

    Publication Year: 1996
    Save to Project icon | Request Permissions | PDF file iconPDF (805 KB)  
    Freely Available from IEEE
  • A comparison of two techniques for detecting seizure in newborn EEG data

    Publication Year: 1996 , Page(s): 3101 - 3104 vol. 6
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (308 KB)  

    This paper considers the problem of automatic classification of newborn electroencephalogram (EEG) data in order to diagnose for seizure. It is shown that good detection performance of seizure EEG is possible using a methodology based on a model for the generation of the EEG. This model is derived from the histology and biophysics of a localised portion of the brain and is thus physically motivated. The model based detection scheme is first presented and used to detect seizure in real newborn EEG data. These results are then compared with an alternative classification approach known as the quadratic detection filter (QDF). It is shown that the model based scheme is far superior to the QDF since it is not adversely affected by the variability (or non-stationarity) of EEG data, which hinders the performance of most traditional EEG classifiers (such as the QDF) View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A soft-decision approach for symbol segmentation within handwritten mathematical expressions

    Publication Year: 1996 , Page(s): 3434 - 3437 vol. 6
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (336 KB)  

    A soft-decision approach for symbol segmentation within on-line sampled handwritten mathematical expressions is presented. Based on stroke-specific features as well as geometrical features between the strokes a symbol hypotheses net is generated. For assistance additional knowledge obtained by a symbol prerecognition stage is used. The results achieved by the segmentation and prerecognition experiments indicate the performance of our approach View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optimized consensus theory

    Publication Year: 1996 , Page(s): 3490 - 3493 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (360 KB)  

    Statistical classification methods based on consensus from several data sources are considered. The methods need weighting mechanisms to control the influence of each data source in the combined classification. The weights are optimized in order to improve the combined classification accuracies. Both linear and non-linear methods are considered for the optimization. A non-linear method which utilizes a neural network is proposed and gives excellent results in experiments. Consensus theory optimized with neural networks outperforms all other methods both in terms of training and test accuracies in the experiments View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • HMM-based handwritten symbol recognition using on-line and off-line features

    Publication Year: 1996 , Page(s): 3438 - 3441 vol. 6
    Cited by:  Papers (8)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (344 KB)  

    This paper addresses the problem of recognizing on-line sampled handwritten symbols. Within the proposed symbol recognition system based on hidden Markov models different kinds of feature extraction algorithms are used analysing on-line features as well as off-line features and combining the classification results. By conducting writer-dependent recognition experiments, it is demonstrated that the recognition rates as well as the reliability of the results is improved by using the proposed recognition system. Furthermore, by applying handwriting data not representing symbols out of the given alphabet, an increase of their rejection rate is obtained View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optoelectronic image processor for multiresolution Gabor filtering

    Publication Year: 1996 , Page(s): 3236 - 3239 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (312 KB)  

    An optoelectronic processor that should allow the implementation of multiresolution Gabor filtering at TV frame rates is described. The system takes the form of a TV camera with modified optics and some simple processing incorporated in the focal-plane detector array. The output of the camera is an array of images, each produced by putting the input image through an angularly-oriented Gabor-type spatial frequency filter and converting the output to baseband. Key elements of the system include: (a) bandpass spatial filtering by pupil modification of an incoherent imaging system, (b) separation of spatial lowpass and spatial bandpass structures by temporal modulation of the bandpass distribution, (c) detection of the temporally modulated distribution by VLSI circuitry in the focal-plane detector array, and (d) multiplexing of multiple spatial bandpass channels in a single smart pixel array View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Properties of a hand-printed Chinese character recognizer based on contextual vector quantization

    Publication Year: 1996 , Page(s): 3486 - 3489 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (272 KB)  

    A hand-printed Chinese character recognizer based on contextual vector quantization (CVQ) has been built previously. In this paper, several properties of the recognizer are discussed. The recognition of 4516 Chinese characters has a successful rate of 91.0%. The output of the recognizer is passed to a language model which when applied to recognize a passage of about 1200 characters raises the rate from 91.5% to 97.5% View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Linear programming in spectral estimation. Application to array processing

    Publication Year: 1996 , Page(s): 3161 - 3164 vol. 6
    Cited by:  Papers (10)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (396 KB)  

    We address the narrow-band source localization problem for arbitrary arrays with known geometry in the presence of arbitrary noise of unknown spatial spectral density. Very few methods are able to handle this problem. We present a very unsophisticated approach whose algorithmic part relies on a standard linear programming algorithm (such as the simplex algorithm available in any scientific program library). The computational complexity of the method is reasonable, the performance appear to be remarkable on simulations. The justification of the procedure and the asymptotic analysis is more complex and much work remains to be done View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Environment-independent continuous speech recognition using neural networks and hidden Markov models

    Publication Year: 1996 , Page(s): 3358 - 3361 vol. 6
    Cited by:  Papers (3)  |  Patents (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (416 KB)  

    Environment-independent continuous speech recognition is important for the successful development of speech recognizers in real world applications. Linear compensation methods do not work well if the mismatches between training; and testing environments are not linear. In this paper, a neural network compensation technique is explored to mitigate the distortion resulting from additive noise, distant-talking, or telephone channels. The advantage of the neural network compensation method is that retraining of a speech recognizer for each particular application is avoided. Furthermore, since neural networks are trained to transform distorted speech feature vectors to those corresponding to clean speech, it may outperform a retrained speech recognizer trained on distorted speech. Three experiments are conducted to evaluate the capability of the neural network compensation method; recognition of additive noisy speech, distant-talking speech, and telephone speech View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Modeling nonlinear systems with cellular neural networks

    Publication Year: 1996 , Page(s): 3513 - 3516 vol. 6
    Cited by:  Papers (10)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (284 KB)  

    A learning procedure for the dynamics of cellular neural networks (CNN) with nonlinear cell interactions is presented. It is applied in order to find the parameters of CNN that model the dynamics of certain nonlinear systems, which are characterized by partial differential equations (PDEs). Values of a solution of the considered PDEs for a particular initial condition are taken as the training pattern at only a small number of points in time. Our results demonstrate that CNN obtained with our method approximate the dynamical behaviour of various nonlinear systems accurately. Results for two nonlinear PDEs, the Φ 4-equation and the sine-Gordon equation, are discussed in detail View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • From characters to words: dynamical segmentation and predictive neural networks

    Publication Year: 1996 , Page(s): 3442 - 3445 vol. 6
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (336 KB)  

    We present the extension of a neural predictive system primitively designed for on-line character recognition to words. Feature extraction is performed after resampling the pen trajectory information, recorded by a digitizing tablet. Each word is modeled by the natural concatenation of letter-models corresponding to the letters composing it. Successive parts of a word trajectory are this way modeled by different neural networks and only transitions from each one to itself or to its right neighbors are permitted. A holistic and dynamical segmentation allows one to adjust letter-models to the great variability of handwriting encountered in the words. Our system combines multilayer neural networks and dynamic programming with an underlying left-right hidden Markov model (HMM). Training was performed on 7000 words from 9 writers, leading to good results in the letter-labelling process, without using any language model View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A unification of relaxation labeling and associative memory

    Publication Year: 1996 , Page(s): 3406 - 3409 vol. 6
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (300 KB)  

    This paper attempts to consolidate the theoretical foundation of the relaxation labeling processes, explore the connections between the relaxation labeling model and the Hopfield associative memory model, and seek their unification. We start by defining a new labeling assignment space and then formulate the relaxation labeling process as a dynamic system of Lyapunov type, which is equipped with a well-defined energy function and described by a naturally fitted updating rule. We present a consistency condition and show that each ω-limit point of the dynamic system gives a consistent labeling. We finally make a peace between the multi-label and one-label relaxation labeling and reveal an interesting result that, for a one-label case, the newly formulated relaxation labeling model reduces to the Hopfield associative memory model View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Multiplierless neural networks for application to digital video

    Publication Year: 1996 , Page(s): 3414 - 3417 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (516 KB)  

    Advances in artificial neural networks have led to several applications such as in computer-vision, image restoration, speech recognition, and pattern classification. However, for widespread practical use these and many other potential applications require large, high-speed networks implemented in efficient custom hardware. This paper investigates an important issue in the implementation of such networks, namely the avoidance of multipliers. Secondly, it evaluates their application in digital video, specifically in high-performance motion prediction and frame reconstruction. The authors discuss the multilayer perceptron and Hopfield neural networks in particular View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Closed-form elliptic location with an arbitrary array topology

    Publication Year: 1996 , Page(s): 3069 - 3072 vol. 6
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (312 KB)  

    An efficient noniterative algorithm for active localization of objects is developed. It is based on intersecting elliptic curves defined by uncertain range-sum measurements between a signal source, the objects, and a number of arbitrarily located receivers. Measurement errors are modelled as being unknown but bounded in amplitude by a closed convex set. Based on this set-theoretic uncertainty model, an error propagation analysis is performed, that allows one to accurately bound estimation errors. For discriminating object primitives and for discarding erroneous measurements, a hypothesis test is derived. The algorithm's computational load is much lower than for grid-based methods and iterative techniques. A recursive formulation is provided to support real-time applications View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • TOM, a new temporal neural net architecture for speech signal processing

    Publication Year: 1996 , Page(s): 3549 - 3552 vol. 6
    Cited by:  Papers (3)  |  Patents (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (344 KB)  

    The neural net model TOM (temporal organization map) that we present in the paper is a new connectionist approach whose time representation is different from the one in classical temporal connectionist models. The architecture is neurobiologically inspired and is dedicated to sensory problems involving a temporal dimension. The basic idea of the TOM model is the propagation of an activity throughout the network whose elements are organized according to a map architecture. This propagation leads to a triggering of a sequence detection. We have applied this new kind of architecture to a spoken digit recognition problem. The results draw near to the results of the best hidden Markov model (HMM) techniques. The interest of such an architecture is its genericity and the possibility to merge several data flows in order to improve the classical performances of neural nets View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A high performance microphone array system for hearing aid applications

    Publication Year: 1996 , Page(s): 3197 - 3200 vol. 6
    Cited by:  Papers (6)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (372 KB)  

    Microphone array technology has been proposed for various audio, teleconference, and hearing aid applications. By forming a focused beam toward the desired speech source, attenuating background noises and rejecting discrete spatial interferers, a microphone array can enhance the SNR/SIR in a noisy environment with significant improvement in speech intelligibility. An array can also perform real time source-localization or direction-of-arrival (DOA) estimation in various applications. We present a high performance prototype PC-based microphone array system for hearing aid applications. Algorithms for maximum energy criterion array weight design needed in the speech processing mode as well as modified broadband near-field MUSIC schemes in the search mode are discussed. Then a PC-based microphone array system using a TMS320C40 DSP is described. Preliminary study of equalizing non-uniform response microphones is also discussed. Finally, some array performance results in free-space and reverberant room conditions are presented View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Variance mismatch: identifying random-test resistance in DSP datapaths

    Publication Year: 1996 , Page(s): 3205 - 3208 vol. 6
    Cited by:  Papers (5)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (428 KB)  

    Pseudorandom built-in self-test (BIST) is an attractive means of testing DSP datapath structures as long as the delay and area overhead can be kept low. One obstacle to low-overhead BIST is random-pattern test-resistant datapath structures. We examine the causes of this resistance in some typical DSP datapaths, and introduce variance mismatch as an analytical tool for identifying these test problems. By understanding the mechanisms behind random-pattern test resistance, it is possible to make design decisions that are compatible with pseudorandom BIST, resulting in reduced test length and higher fault coverage. Variance matching theory is applied to the design of two large FIR filters, resulting in an 88% reduction in the number of missed faults for a fixed test length, and two orders-of-magnitude reduction in test length for a specific fault coverage target View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Iterative maximum-likelihood/cross-correlation algorithm for reflection and pulse TOA estimation

    Publication Year: 1996 , Page(s): 3133 - 3136 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (296 KB)  

    An iterative method that combines a parametric channel model with the maximum likelihood and cross-correlation estimation techniques has been developed to estimate the reflection, creeping wave, and times of arrival (TOAs) from an echo (backscatter) emanating from a rigid acoustic sphere. The modelled echo is assumed to be more complex than a linear combination of delayed and scaled versions of the transmitted pulse, and the individual backscatter pulses need not be disjoint. Experimental results demonstrate that the algorithm is successful in determining the proper time delays of the reflection and the creeping wave for a variety of received signal-to-noise ratios (SNRs) View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Gerschgorin radii based source number detection for closely spaced signals

    Publication Year: 1996 , Page(s): 3053 - 3056 vol. 6
    Cited by:  Papers (2)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (312 KB)  

    We first analyze the resolution performance of the Gerschgorin radii based source number detector proposed by Wu, Yang and Chen (see IEEE Trans. SP, vol.43, no.6, p.1325-33, 1995) for independent closely-spaced plane waves. Based upon the analysis, we can easily verify the resolution threshold of the Gerschgorin radii based algorithm. For improvement of the detection performance, we then further propose a weighted Gerschgorin radii algorithm. With closed-form expressions and simulations, we find that the analyzed performance in terms of the number of sensors, the relative angular separation of emitters, and the signal-to-noise ratios can properly characterize their behaviour for source number detection View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Fast nonlinear adaptive filtering using a partial window conjugate gradient algorithm

    Publication Year: 1996 , Page(s): 3541 - 3544 vol. 6
    Cited by:  Papers (3)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (288 KB)  

    In this paper a modified form of the partial conjugate gradient algorithm is presented for use in nonlinear filtering using neural networks. The algorithm is based on using a gradient average window to provide a trade-off between convergence rate and complexity which, depending on the choice of averaging window, is (in both complexity and speed of convergence) intermediate between the conventional backpropagation (BP) algorithm and the Newton methods. An additional simplification is introduced by replacing the calculated optimum step size αk by a normalized step size α¯, in the same manner as the normalized LMS algorithm. This new algorithm is applied to a cascaded neural network/nonlinear least mean squares structure for the identification of a nonlinear system. This proposed algorithm demonstrates improved convergence rates with even small choices of window size View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Adaptive nonlinear system identification using minimal radial basis function neural networks

    Publication Year: 1996 , Page(s): 3521 - 3524 vol. 6
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (344 KB)  

    In this paper, an adaptive identification scheme for nonlinear systems using a minimal radial basis function neural network (RBFNN) is presented. This scheme combines the growth criterion of the resource-allocating network (RAN) of Platt (1991) with a pruning strategy based on the relative contribution of each hidden unit to the overall network output. While being applied to nonlinear system identification, this approach enables the number of hidden layer neurons in the network to be adjusted to the changing system dynamics, the resulting neural network also leads to a minimal topology for the RBFNN. Simulations are carried out to recursively identify two nonlinear systems with time-varying dynamics. The performance of the proposed algorithm is compared with the recursive hybrid algorithm for system identification proposed by Chen et al. (1992). The proposed algorithm in this paper is shown to realize a RBFNN with far fewer hidden neurons and better accuracy View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Wreath products for image processing

    Publication Year: 1996 , Page(s): 3581 - 3584 vol. 6
    Cited by:  Papers (4)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (484 KB)  

    We present a wreath product approach for matched filtering to detect rotated copies of a template in an image. We view the image as a homogeneous space for a wreath product, a noncommutative symmetry group. The corresponding Fourier analysis has a natural multiresolution structure and accompanying efficient algorithm which we explain and illustrate with an example. The associated matched filter is a new example of the use of a noncommutative convolution for image processing. Numerical experiments are described in which this noncommutative approach outperforms standard Fourier-based methods View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Chinese all syllables recognition using combination of multiple classifiers

    Publication Year: 1996 , Page(s): 3494 - 3497 vol. 6
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (364 KB)  

    Chinese all syllables recognition is described. Chinese all syllables recognition is divided into base syllable recognition disregarding the tones and 4 tones recognition. For base syllable recognition, we used a combination of two multisegment vector quantization (MSVQ) classifiers based on different features (instantaneous and transitional features of speech). For the tones recognition, the vector quantization (VQ) classifier is first used, and is comparable to a multilayer perceptron (MLP) classifier. Next, a combination of a distortion based classifier (VQ) and a discriminant based classifier (MLP) is proposed. An evaluation has been carried out using the standard Chinese syllable database CRDB, and experimental results have shown that the combined classifiers can improve the recognition performance. The recognition accuracy for base syllable and tones is 96.48% and 99.82% respectively View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Energy demodulation of two-component AM-FM signals with application to speaker separation

    Publication Year: 1996 , Page(s): 3517 - 3520 vol. 6
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (340 KB)  

    In this paper, an efficient low-complexity algorithm is presented for the separation and demodulation of two-component AM-FM signals and applied to the separation of voice-modulated FM signals. The proposed algorithm is based on the generating differential equation of the mixture signal and nonlinear differential energy operators View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A new data reduction algorithm for pattern classification

    Publication Year: 1996 , Page(s): 3446 - 3449 vol. 6
    Cited by:  Papers (1)
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (264 KB)  

    The design of pattern classifiers such as multiprototype classifiers and neural network classifiers such as learning vector quantization and radial basis function neural networks requires reducing the size of the training data sets. In addition, memory storage, computation complexity and time, and data redundancy demand many pattern classifiers to use a smaller subset of a training data set. In this paper, we present a data reduction algorithm which automatically selects the subset of training data that faithfully represents the training data set for pattern classification. The applicability of this algorithm is demonstrated through k-nearest neighbor and learning vector quantization neural networks classifiers using both speech and synthetic data sets View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.