Autoassociative Neural Network Models Research Articles

Speaker change detection in casual conversations using excitation source features

In this paper we propose a method for speaker change detection using features of excitation source of the speech production mechanism. The method uses neural network models to capture the speaker-specific information from a signal that represents predominantly the excitation source. The focus in this paper is on speaker change detection in casual telephone conversations, in which short (<5 s) speaker turns are common. Excitation source features are a better choice for modeling a speaker, when limited amount of speech data is available, when compared to the vocal tract system features. Linear prediction residual is used as an estimate of the excitation source signal. Autoassociative neural network models are proposed to capture the higher order relations among the samples of the residual signal. Speaker models are generated for every one second of voiced speech from the first few seconds of the conversation. These models are used to detect the speaker change points. Performance of the proposed method for speaker change detection is evaluated on a database containing several two-speaker conversations.

Face Verification Using Template Matching

Human faces are similar in structure with minor differences from person to person. These minor differences may average out while trying to synthesize the face image of a given person, or while building a model of face image in automatic face recognition. In this paper, we propose a template-matching approach for face verification, which neither synthesizes the face image nor builds a model of the face image. Template matching is performed using an edginess-based representation of the face image. The edginess-based representation of face images is computed using 1-D processing of images. An approach is proposed based on autoassociative neural network models to verify the identity of a person. The issues of pose and illumination in face verification are addressed.

Significance of image representation for face verification

In this paper we discuss the significance of representation of images for face verification. We consider three different representations, namely, edge gradient, edge orientation and potential field derived from the edge gradient. These representations are examined in the context of face verification using a specific type of correlation filter, called the minimum average correlation energy (MACE) filter. The different representations are derived using one-dimensional (1-D) processing of image. The 1-D processing provides multiple partial evidences for a given face image, one evidence for each direction of the 1-D processing. Separate MACE filters are used for deriving each partial evidence. We propose a method to combine the partial evidences obtained for each representation using an auto-associative neural network (AANN) model, to arrive at a decision for face verification. Results show that the performance of the system using potential field representation is better than that using the edge gradient representation or the edge orientation representation. Also, the potential field representation derived from the edge gradient is observed to be less sensitive to variation in illumination compared to the gray level representation of images.

Multimodal person authentication using speech, face and visual speech

This paper presents a method for automatic multimodal person authentication using speech, face and visual speech modalities. The proposed method uses the motion information to localize the face region, and the face region is processed in YC r C b color space to determine the locations of the eyes. The system models the nonlip region of the face using a Gaussian distribution, and it is used to estimate the center of the mouth. Facial and visual speech features are extracted using multiscale morphological erosion and dilation operations, respectively. The facial features are extracted relative to the locations of the eyes, and visual speech features are extracted relative to the locations of the eyes and mouth. Acoustic features are derived from the speech signal, and are represented by weighted linear prediction cepstral coefficients (WLPCC). Autoassociative neural network (AANN) models are used to capture the distribution of the extracted acoustic, facial and visual speech features. The evidence from speech, face and visual speech models are combined using a weighting rule, and the result is used to accept or reject the identity claim of the subject. The performance of the system is evaluated for newsreaders in TV broadcast news data, and the system achieves an equal error rate (EER) of about 0.45% for 50 subjects.

Modelling adaptation aftereffects in associative memory

We present a simple autoassociative neural network model of cortical adaptation that replicates a high-level category aftereffect. We study in our model the effect of a priming stimulus on retrieval from a briefly presented partial cue. We also explore, in our general framework, which learning conditions permit correct pattern retrieval.

Extraction of speaker-specific excitation information from linear prediction residual of speech

In this paper, through different experimental studies we demonstrate that the excitation component of speech can be exploited for speaker recognition studies. Linear prediction (LP) residual is used as a representation of excitation information in speech. The speaker-specific information in the excitation of voiced speech is captured using the AutoAssociative Neural Network (AANN) models. The decrease in the error during training and recognizing correct speakers during testing demonstrates that the excitation component of speech contains speaker-specific information and is indeed being captured by the AANN models. The study on the effect of different LP orders demonstrates that for a speech signal sampled at 8 kHz, the LP residual extracted using LP order in the range 8–20 best represents the speaker-specific excitation information. It is also demonstrated that the proposed speaker recognition system using excitation information and AANN models requires significantly less amount of data both during training as well as testing, compared to the speaker recognition system using vocal tract information. Finally the speaker recognition studies on NIST 2002 database demonstrates that even though, the recognition performance from the excitation information alone is poor, when combined with evidence from vocal tract information, there is significant improvement in the performance. This result demonstrates the complementary nature of the excitation component of speech.

Comparative study on state analysis of BOP in NPPs

Under the deregulation environment, the performance enhancement of balance of plant (BOP) or the turbine cycle in nuclear power plants is being highlighted. To analyze performance level of the turbine cycle, we use the performance test procedures provided from an authorized institution such as American Society of Mechanical Engineers. However, through plant investigation, it was proved that the requirements of the performance test procedures about the reliability and quantity of sensors was difficult to be satisfied. As a solution of this weakness, a state analysis method that is the expanded concept of signal validation, was proposed on the basis of the statistical and the neural network approaches. The authors recommended the statistical linear regression model by analyzing correlation among turbine cycle parameters as a reference state analysis method. Its advantage is that its derivation is not heuristic, it is possible to calculate model uncertainty, and it is easy to apply to an actual plant. The error of the statistical linear regression model is below 3% under normal as well as abnormal system states. Additionally the auto-associative neural network (AANN) model was recommended since the statistical model is impossible to apply to the validation of all of the sensors and is sensitive to the outlier that is the signal located out of statistical distribution. Because there are a lot of sensors need to be validated in turbine cycle, principal component analysis (PCA) and wavelet analysis (WA) were applied as a pre-processor for the reduction of input dimension and for the enhancement of training accuracy, and the results were compared. Both are effective to reduce the number of the input and output dimension, but WA is superior to PCA in the compensation of the outliers. The outlier localization capability of WA enhanced the robustness of the AANN. In case of the AANN with WA, the training accuracy of the model was nearly 100% and the trained neural network restored the degraded signals to the values within /spl plusmn/3% of the true signals.

Classification of fermentation performance by multivariate analysis based on mean hypothesis testing

Multivariate analysis, such as principal component analysis and artificial autoassociative neural networks, is currently extensively applied to feature capturing, physiological state recognition, fault detection and bioprocess control. However, it is not clear which process variable should be selected as an important input for multivariate analysis to analyze physiological conditions and/or bioprocess performance a priori. An efficacious method to select more informative process variables from the repository of historical data is highly desired. In this study, we focused on a premodeling step. Mean hypothesis testing (MHT) was used to select appropriate variables for multivariate analysis. Fermentation data sets were classified into two classes “good” and “bad” according to the MHT results. The results showed that selecting discriminating process variables from the historical database by MHT enhanced the overall effectiveness of multivariate analysis prior to principal component analysis and artificial autoassociative neural network model creation.

Data preprocessing and output evaluation of an autoassociative neural network model for online fault detection in virginiamycin production

In this study, an artificial autoassociative neural network (AANN) was used online to detect deviations from normal antibiotic production fermentation using conventional process variables. To improve the efficiency of extracting hidden information contained in multidimensional process variables, and to finally render the AANN adequate for fault detection, we explored the following methods: selection of process variables; preprocessing of data that involved normalizing the training data of the AANN; and evaluation of data that involved assessing the output of the AANN. A method for fault detection in virginiamycin M and S production by Streptomyces virginiae was successfully developed based on these techniques.

AANN: an alternative to GMM for pattern recognition

The objective in any pattern recognition problem is to capture the characteristics common to each class from feature vectors of the training data. While Gaussian mixture models appear to be general enough to characterize the distribution of the given data, the model is constrained by the fact that the shape of the components of the distribution is assumed to be Gaussian, and the number of mixtures are fixed a priori. In this context, we investigate the potential of non-linear models such as autoassociative neural network (AANN) models, which perform identity mapping of the input space. We show that the training error surface realized by the neural network model in the feature space is useful to study the characteristics of the distribution of the input data. We also propose a method of obtaining an error surface to match the distribution of the given data. The distribution capturing ability of AANN models is illustrated in the context of speaker verification.

Classification of Fermentation Performance by Multivariate Analysis Based on Mean Hypothesis Testing

Multivariate analysis, such as principal component analysis and artificial autoassociative neural networks, is currently extensively applied to feature capturing, physiological state recognition, fault detection and bioprocess control. However, it is not clear which process variable should be selected as an important input for multivariate analysis to analyze physiological conditions and/or bioprocess performance a priori. An efficacious method to select more informative process variables from the repository of historical data is highly desired. In this study, we focused on a premodeling step. Mean hypothesis testing (MHT) was used to select appropriate variables for multivariate analysis. Fermentation data sets were classified into two classes "good" and "bad" according to the MHT results. The results showed that selecting discriminating process variables from the historical database by MHT enhanced the overall effectiveness of multivariate analysis prior to principal component analysis and artificial autoassociative neural network model creation.

Dynamics of an auto-associative neural network model with arbitrary connectivity and noise in the threshold

Dynamics of an auto-associative neural network model with arbitrary connectivity and noise in the threshold

Dynamics of an auto-associative neural network model with arbitrary connectivity and noise in the threshold

The method of statistical neurodynamics is used to analyse retrieval dynamics of an auto-associative neural network model. The model has arbitrarily specified connectivity and static noises are added to threshold values. The method is based only on probability and approximation calculations. Connections between neurons are determined by a version of the Hebb rule (correlation-type rule), and some of them are removed at random. It is shown that the capacity of the network per connection is a monotone decreasing function of connectivity if there are no noises in the threshold. When there are noises in the threshold there exists an optimal value of the sparsity of connections which yields the maximum capacity for a fixed noise level. In addition, effects of systematic removal of connections in contrast with random removal, i.e. structured models, are discussed. It is shown that a wide range of neural network models, such as a bidirectional associative memory network or a layered network, are special cases of...