A Comparative Study of Isolated Word Recognizer Using SVM and WaveNet
01 Jan 2018-pp 139-147
TL;DR: Speaker-independent isolated word recognition system is proposed using the Mel-Frequency Cepstral Coefficients feature extraction method to create the feature vector and the results are compared in terms of the maximum accuracy obtained and the number of iterations taken to achieve this.
Abstract: In this paper, speaker-independent isolated word recognition system is proposed using the Mel-Frequency Cepstral Coefficients feature extraction method to create the feature vector. Support vector machine, sigmoid neural net, and the novel wavelet neural network are used as classifiers and the results are compared in terms of the maximum accuracy obtained and the number of iterations taken to achieve this. The effect of stretch factor on the accuracy of classification for WaveNets is shown in the results. The number of features is also varied using dimension reduction technique and its effect on the accuracies is studied. The data is prepared using feature scaling and dimensionality reduction before training SVM and NN classifiers.
Citations
More filters
18 Jul 2022
TL;DR: In this article , an online incremental growing neural gas (oiSGNG) with a feature extractor is proposed as a voice command classifier for low-cost implementations, such as an electric wheelchair.
Abstract: Online learning approach allows an Artificial Neural Network (ANN) to solve dynamic real-world problems. In this context, the objective of this work is to implement ANN-based voice recognition models with focus on class-incremental learning in real time, for low-cost implementations, such as an electric wheelchair. In this paper, the online incremental Supervised Growing Neural Gas (oiSGNG) with a feature extractor is proposed as a voice command classifier. About this model, two contributions are presented: (i) nodes are inserted according to an exponential function, that results in a higher accuracy rate with fewer nodes, which implies less latency; (ii) adaptive oiSGNG, this model is a novel implementation that enables online learning. In offline experiments, the model proposed performs better than the Self-Organizing Map (SOM) in its topological and supervised version. After simulations and experiments, it is proposed to use a keyword to avoid false positives. In the results, the accuracy of the proposed model is better than the original oiSGNG.
18 Jul 2022
TL;DR: The objective of this work is to implement ANN-based voice recognition models with focus on class-incremental learning in real time, for low-cost implementations, such as an electric wheelchair.
Abstract: Online learning approach allows an Artificial Neural Network (ANN) to solve dynamic real-world problems. In this context, the objective of this work is to implement ANN-based voice recognition models with focus on class-incremental learning in real time, for low-cost implementations, such as an electric wheelchair. In this paper, the online incremental Supervised Growing Neural Gas (oiSGNG) with a feature extractor is proposed as a voice command classifier. About this model, two contributions are presented: (i) nodes are inserted according to an exponential function, that results in a higher accuracy rate with fewer nodes, which implies less latency; (ii) adaptive oiSGNG, this model is a novel implementation that enables online learning. In offline experiments, the model proposed performs better than the Self-Organizing Map (SOM) in its topological and supervised version. After simulations and experiments, it is proposed to use a keyword to avoid false positives. In the results, the accuracy of the proposed model is better than the original oiSGNG.
References
More filters
Journal Article•
TL;DR: Different types of basis functions are used as the activation function in the hidden nodes of the radial basis function network and the wavelet neural network in estimating periodic, exponential and piecewise continuous functions.
Abstract: Function approximation, which finds the underlying relationship from a given finite input-output data is the fundamental problem in a vast majority of real world applications, such as prediction, pattern recognition, data mining and classification. Various methods have been developed to address this problem, where one of them is by using artificial neural networks. In this paper, the radial basis function network and the wavelet neural network are applied in estimating periodic, exponential and piecewise continuous functions. Different types of basis functions are used as the activation function in the hidden nodes of the radial basis function network and the wavelet neural network. The performance is compared by using the normalized square root mean square error function as the indicator of the accuracy of these neural network models.
83 citations
TL;DR: A wavelet mother function selection algorithm with minimum mean squared error is proposed and MWFWNN network, a variant of wavelet neural networks, has a good predictive ability and can quickly and accurately complete target threat assessment.
Abstract: Target threat assessment is a key issue in the collaborative attack. To improve the accuracy and usefulness of target threat assessment in the aerial combat, we propose a variant of wavelet neural networks, MWFWNN network, to solve threat assessment. How to select the appropriate wavelet function is difficult when constructing wavelet neural network. This paper proposes a wavelet mother function selection algorithm with minimum mean squared error and then constructs MWFWNN network using the above algorithm. Firstly, it needs to establish wavelet function library; secondly, wavelet neural network is constructed with each wavelet mother function in the library and wavelet function parameters and the network weights are updated according to the relevant modifying formula. The constructed wavelet neural network is detected with training set, and then optimal wavelet function with minimum mean squared error is chosen to build MWFWNN network. Experimental results show that the mean squared error is 1.23 × 10(-3), which is better than WNN, BP, and PSO_SVM. Target threat assessment model based on the MWFWNN has a good predictive ability, so it can quickly and accurately complete target threat assessment.
69 citations
12 May 2014
TL;DR: This paper investigates the use of convolutional and fully-connected neural networks with different activation functions (sigmoid, rectified linear, and maxout) for distant speech recognition of meetings recorded using microphone arrays, and indicates that neural network models are capable of significant improvements in accuracy compared with discriminatively trained Gaussian mixture models.
Abstract: Distant conversational speech recognition is challenging owing to the presence of multiple, overlapping talkers, additional non-speech acoustic sources, and the effects of reverberation. In this paper we review work on distant speech recognition, with an emphasis on approaches which combine multichannel signal processing with acoustic modelling, and investigate the use of hybrid neural network / hidden Markov model acoustic models for distant speech recognition of meetings recorded using microphone arrays. In particular we investigate the use of convolutional and fully-connected neural networks with different activation functions (sigmoid, rectified linear, and maxout). We performed experiments on the AMI and ICSI meeting corpora, with results indicating that neural network models are capable of significant improvements in accuracy compared with discriminatively trained Gaussian mixture models.
51 citations
Proceedings Article•
01 Sep 2006TL;DR: A pure SVM-based continuous speech recognizer, using the SVM to make decisions at frame-level, and a Token Passing algorithm to obtain the chain of recognized words.
Abstract: Although Support Vector Machines (SVMs) have been proved to be very powerful classifiers, they still have some problems which make difficult their application to speech recognition, and most of the tries to do it are combined HMM-SVM solutions. In this paper we show a pure SVM-based continuous speech recognizer, using the SVM to make decisions at frame-level, and a Token Passing algorithm to obtain the chain of recognized words. We consider a connected digit recognition task with both, digits themselves and number of digits, unknown. The experimental results show that, although not yet practical due to computational cost, such a system can get better recognition rates than traditional HMM-based systems (96.96% vs. 96.47%). To overcome computational problems, some techniques as the Mega-GSVCs can be used in the future.
44 citations
Proceedings Article•
29 Dec 2007TL;DR: In this paper, the radial basis function network and the wavelet neural network are applied in estimating periodic, exponential and piecewise continuous functions, and the performance is compared by using the normalized square root mean square error function as the indicator of the accuracy of these neural network models.
Abstract: Function approximation, which finds the underlying relationship from a given finite input-output data is the fundamental problem in a vast majority of real world applications, such as prediction, pattern recognition, data mining and classification. Various methods have been developed to address this problem, where one of them is by using artificial neural networks. In this paper, the radial basis function network and the wavelet neural network are applied in estimating periodic, exponential and piecewise continuous functions. Different types of basis functions are used as the activation function in the hidden nodes of the radial basis function network and the wavelet neural network. The performance is compared by using the normalized square root mean square error function as the indicator of the accuracy of these neural network models.
40 citations