Malay speaker identification using Neural Networks

This paper investigates the use of Neural Networks in recognizing Malay vowels of children in speakerindependent manner. Malay vowels are comprised of /a/, /e/, / Open image in new window /, /i/, /o/ and /u/. Speech database is collected from 300 Malay children between seven and twelve years old. Each speaker contributes two samples per vowel sound. The speech database is organized equally into training set and test set. The speech sounds are sampled at 20 kHz with 16 bit resolution. A single frame of cepstral coefficients is extracted around the vowel onset point using Linear Predictive Coding. Multi-Layer Perceptron (MLP) with one hidden-layer is used to train and recognize the vowel sounds. The output of the MLP consists of 6 neurons, which correspond to the 6 vowel sounds. Experiments are conducted to determine the optimal signal length of vowels, and hidden neuron number of MLP. A maximum recognition rate of 75.00% is achieved at signal length of 30ms and 35ms.

Most of the speech recognitions are based on adult speech sounds. Less research is done in the recognition of children speech sounds. The speech of children is more dynamic and inconsistent if compared to adult's speech. This paper investigates the use of neural networks in recognizing 6 Malay vowels of Malay children in a speaker-independent manner. Multi-layer perceptron with one hidden layer was used to recognize these vowels. The multi-layer perceptron was trained and tested with speech samples of Malay children with their ages between seven and ten years old. A single frame of cepstral coefficients was extracted around the vowel onset point using linear predictive coding. The vowel length was examined from 5 ms to 70 ms. Experiments were conducted to determine the optimal vowel length as well as the number of cepstral coefficients.

The aim of the study was to inspect the acoustic properties and sleep characteristics of a preapneic snoring sound. The feasibility of forecasting upcoming respiratory events by snoring sound was also investigated. Participants with habitual snoring or a heavy breathing sound during sleep were recruited consecutively. Polysomnography was conducted, and snoring-related breathing sound was recorded simultaneously. Acoustic features and sleep features were extracted from 30-second samples, and a machine learning algorithm was used to establish 2 prediction models. A total of 74 eligible participants were included. Model 1, tested by 5-fold cross-validation, achieved an accuracy of 0.92 and an area under the curve of 0.94 for respiratory event prediction. Model 2, with acoustic features and sleep information tested by Leave-One-Out cross-validation, had an accuracy of 0.78 and an area under the curve of 0.80. Sleep position was found to be the most important among all sleep features contributing to the performance of the 2 models. Preapneic sound presented unique acoustic characteristics, and snoring-related breathing sound could be deployed as a real-time apneic event predictor. The models, combined with sleep information, serve as a promising tool for an early warning system to forecast apneic events. Wang B, Yi X, Gao J, etal. Real-time prediction of upcoming respiratory events via machine learning using snoring sound signal. J Clin Sleep Med. 2021;17(9):1777-1784.

A cascade formant model is well suited to describe certain speech segments, such as vowels and vowel-like sounds. The formant model is also useful because the relationship between formants and the vocal tract configurations are well understood; however, this model is not adequate for other speech sounds, such as stops, fricatives, nasals, etc. On the other hand, LPC (linear predictive coding) analysis, of sufficiently high order, can adequately describe the spectrum of any speech sound, but the relationship between the LPC parameters and the spectrum or vocal tract configuration is not obvious. This paper describes a speech analysis/synthesis scheme that uses both formants and LPC parameters for different sections of a speech signal. Thus, in some regions, the benefit of the formant model can be utilized, while in other regions, the LPC representation can be used to obtain a good description of the speech spectrum. The analysis algorithm resolves the problem of discontinuities that arise from using the two different spectral representations. When provided with the same multipulse signal, the method of speech analysis described in this paper produces resynthesized speech of the quality of multipulse LPC.

In recent scenario speaker recognition is used in many places for security purpose. Speaker recognition is technique which can verify or identify the person who is speaking. It is different than speech recognition system. In this paper we have discussed the feature extraction techniques like Mel Frequency Cepstral Co-efficient (MFCC), Linear Predictive Coding (LPC) etc. and classification techniques like Gaussian Mixture Model (GMM), Artificial Neural Networks (ANN) etc. that is available for speaker recognition. Also the survey is about how to get better efficiency in terms of speaker recognition rate by simply modifying the existing feature extraction and classification techniques.

Purpose. At the present stage, network attack detection systems based on the following neural networks are most often offered: multilayer perceptron, Kohonen network or self-organizing map and their combinations. The efficiency problem of two approaches to detect attacks on a computer network using neural network technology based on the normalized data of the open NSL-KDD database is considered. Methodology. As an architectural solution to the network attack detection system, it is proposed to consider the following approaches: based on one neural network determining the attack class (first approach) and an ensemble of five neural networks (second approach), which at the first stage determines the attack category (DoS, Probe, U2R , R2L), and in the second stage, the attack class belonging to a certain category. Findings. Based on the neural networks created in the MatLAB program, a study was conducted of their error on the length of the training sample using various training algorithms: Levenberg-Marquardt; Bayesian Regularization; Scaled Conjugate Gradient with different numbers of hidden neurons (minimum, average and maximum). Certain optimal parameters of neural networks with two approaches were determined. Originality. In the course of conducting experiments with various approaches, the results obtained were: TP (True Positive); FP (False Positive); FN (False Negative); TN (True Negative), based on which the following indicators were calculated for assessing the quality of solutions: correct determination of network attacks; false positives; reliability; accuracy and completeness, which prove the feasibility of using an ensemble of neural networks (second approach). Practical value. On the created neural networks with various approaches, studies were conducted: the operating time of neural networks; errors of the first kind; errors of the second kind. According to the results of the first study, the average operating time of an ensemble of neural networks is 0.92 s, and the operating time of a neural network (according to the first approach) is 2.21 s. According to the results of the second study, the error of the first kind using an ensemble of neural networks is 2.17%, and using the neural network (the first approach) – 7.39%. According to the results of the third study, the error of the second kind using an ensemble of neural networks is 3.91%, and using the neural network (the first approach) – 6.96%, which is confirmed by the efficiency of using an ensemble of neural networks (second approach).

The objective of this study is to show how a multi-layer perceptron (MLP) neural network can be used to model a CMM measuring process. To date, most MLP-based process models have been established for process mean only. An innovative approach is proposed to model simultaneously the mean and the variation of a CMM process using one integrated MLP architecture. Therefore, the MLP-based model obtained captures not only the process mean but also the process variation information. Selected issues related to neural network training are also discussed. Specifically, the guideline that was proposed by Mirchandani and Cao (1989) for selecting a number of hidden neurons is tested to determine the effects of the number of hidden neurons. The performances of two different learning algorithms - back-propagation with momentum factor (BPM) and the Fletcher-Reeves (FR) algorithm - are studied in terms of CPU time, training error, and generalization error.

The paper is focused on the idea to demonstrate the advantages of deep learning approaches over ordinary shallow neural network on their comparative applications to image classifying from such popular benchmark databases as FERET and MNIST. An autoassociative neural network is used as a standalone program realized the nonlinear principal component analysis for prior extracting the most informative features of input data for neural networks to be compared further as classifiers. A special study of the optimal choice of activation function and the normalization transformation of input data allows to improve efficiency of the autoassociative program. One more study devoted to denoising properties of this program demonstrates its high efficiency even on noisy data. Three types of neural networks are compared: feed-forward neural net with one hidden layer, deep network with several hidden layers and deep belief network with several pretraining layers realized restricted Boltzmann machine. The number of hidden layer and the number of hidden neurons in them were chosen by cross-validation procedure to keep balance between number of layers and hidden neurons and classification efficiency. Results of our comparative study demonstrate the undoubted advantage of deep networks, as well as denoising power of autoencoders. In our work we use both multiprocessor graphic card and cloud services to speed up our calculations. The paper is oriented to specialists in concrete fields of scientific or experimental applications, who have already some knowledge about artificial neural networks, probability theory and numerical methods.

Speaker recognition is a biometric technique which uses individual speakers voice samples as a input for recognition purpose. Main goal of this work is to obtain better accuracy for speaker recognition system for large number of voice database. In this paper, a comparative study is made between various combinations of features for speaker identification system with feedforward artificial neural network(FFANN) and support vector machine for 10 and 20 speakers. Linear predictive coding, Mel frequency cepstral coefficient and zero crossing rate are used as a feature extraction techniques. Each features are tested separately and in combination with FFANN and SVM classifier on Matlab software. For ANN classifier 70% of total database are used for training,15% for validation and remaining 15% data are used for testing, number of hidden layers and number of neurons used are 2 and 80 respectively. It is observed from the result that efficiency of the system is not dropping by increasing number of speakers from 10(320 voice samples) to 20(640 voice samples) when using combination of LPC, MFCC, ZCR features with ANN classifier.

Part 1 Vision: image feature location in multi-resolution images using a hierarchy of multi-layer perceptrons, J. Vincent et al training multi-layer perceptrons for facial feature location - a case study, Jon Waite the detection of eyes in facial images using radial basis functions, R. Debenham and S. Garth training and testing of neural net window operators on spatiotemporal image sequences, M. Wright a neural network feature detector using a multi-resolution pyramid, C. Hand et al image classification using Gabor representations with a neural net, P. Lisboa. Part 2 Speech: spoken alphabet recognition using multi-layer perceptrons, Philip Woodland spoken independent vowel recognition, L. Smith and C. Tang dissection of perceptron structures in speech and speaker recognition, J. Mason and E. Andrews segmental sub-word unit classification using a multi-layer perceptron, S. Smyth. Part 3 Natural language: a single layer higher order neural net and its applications to context free grammar recognition, P. Wyard and C. Nightingale functional compositionality and soft preference rules, N. Sharkey applications of multi-layer perceptrons in text-to-speech synthesis, W. Ainsworth and N. Warren. Part 4 Implementation: finite word length, integer arithmetic multi-layer perceptron modelling for hardware realization, J. Vincent a VSLI architecture for implementing neural networks with on-chip propagation algorithm learning, D. Myers et al an opto-electronic neural network processor, N. Barnes et al. Part 5 Architectures: a dynamic topology net, C. Nightingale the stochastic search network, J. Bishop and P. Torr node sequence networks, R. Linggard some dynamical properties of neural networks, S. Olafsson.

This paper examines the performance improvement via combining feature vectors for speaker verification. As feature vectors, we use LPC (linear predictive coding) spectral envelope (LPCSE), Mel-frequency cepstral coefficients (MFCC) and piecewise LPC (pLPC) pole distribution (pLPCPD). Here, pLPCPD is the feature which we have introduced and developed, and this paper shows a relationship of the underlying pLPC speech signal model for speaker verification with ARMA and AR models for speech production. We execute experiments of speaker verification with the combinations of the features. We examine the experimental results and show that (1) the smallest AER (average error rate) is achieved by MFCC followed in order by pLPCPD and LPCSE, (2) the AER achieved by LPCSE, MFCC and LPCSE+MFCC (combination of LPCSE and MFCC) is reduced by combining pLPCPD, (3) there are conflicting features in LPCSE, MFCC and pLPCPD to increase the AER, and the smallest AER is obtained by MFCC+pLPCPD among the combinations of LPCSE, MFCC and pLPCPD. We also investigate the relationship between pLPCPD and the aperiodicity used for vocal cord excitation in recent vocoders, which seems to explain the above result (2).

Efficient functioning is the first and foremost problem of the control system of technological objects while evaluating its performance. The evaluation criteria can be as follows: the degree of relation between the control action and the control signal; the integrated index of the system dynamics; the statistical measure of the system quality. The article suggests an approach towards forecasting the efficiency of the control system based on neural networks. The efficiency forecasting of the control system on the basis of neural networks with a various number of hidden neurons has been performed. The best result has been demonstrated by the 4-7-2 MLP (multilayer perceptron) network which possesses four input, seven hidden and two output neurons respectively. The MLP has secured 99.19 % performance in the training sample and 98.97 % in the testing sample. The results obtained secure the rational use of the chosen neural network architecture for the purpose of forecasting the efficiency of the control system.

Automated model generation (AMG) is a systematic and efficient method to develop an artificial neural network (ANN) model for microwave components. By incorporating the Bayesian theory into AMG, the most compact multilayer perceptron (MLP) model with the highest accuracy can be developed. However, the existing Bayesian-based AMG method is only suitable for single-hidden-layer MLP. In this letter, we propose a novel Bayesian-assisted AMG method for neural networks with multiple hidden layers. A systematic algorithm is proposed to determine the optimal number of hidden layers and the number of hidden neurons in each hidden layer. Using the proposed method, a compact and accurate MLP model with a multi-hidden layer structure can be systematically developed. The proposed method can achieve a higher model accuracy than the previous Bayesian-based method with almost the same number of model network parameters. Two microwave filters are used to demonstrate the advantages of the proposed method.

Electric Powered Wheelchair (EPW) used by people with disabilities, who cannot move or with a limited ability to move. The traditional technique that used to control the powered wheelchair can be a challenging and stressful operation. Therefore, there is a demand to find alternative ways to perform this process. In that, the voice recognition is a natural form of pointing that can use to replace the joystick whilst still allowing for similar control. Neural network one of the effective ways to recognize the speech commands, because of its ability to grant a trainable powered wheelchair controller with a convenient way, ease of use and attractiveness. In this situation, the neural networks are providing a trainable system for each person irrespectively of his disability. Smart Electric-Powered Wheelchair (SEPW) is an electric-powered wheelchair based on Artificial Neural Networks (ANN) Technique can recognize the voice commands and replace the traditional control of powered wheelchairs with intelligent technology. So, a neural network for the feed-forward Multi-layer perceptron has been trained to recognize isolated spoken words (commands), such as ‘right, ‘left,' ‘stop' and so on. Initially, the design of the system's network started with one hidden layer using five neurons (according to [1] formula). During the training process, the number of hidden neurons has been increased to keep the mean square error (MSE) of the system minimum as possible. Furthermore, the system structure is designed based on a Multi Layers Perceptron (MLP) neural system for five output neurons, which represent the powered wheelchair's commands

An artificial neural network (ANN) is applied to perform the task of acoustic-to-articulatory inversion. The objective is to model the highly nonlinear mapping from linear predictive coding (LPC) code to corresponding articulatory parameters with a multilayer perceptron ANN structure. Such information will facilitate the study of the relationships between the acoustic signal and the physical vocal tract which produces it. Several novel approaches for devising the ANN structure have been evaluated. Specifically, the performance of two learning algorithms, a backpropagation (BP) algorithm, and a random optimization (RM) algorithms, are compared. To reduce excessive, redundant hidden units in the multilayer perceptron model, a singular value decomposition is applied to either the weight matrix or the output covariance matrix of the hidden units to check their corresponding ranks. In both cases, their ranks are closely related to the number of essential decision regions in the input data. >