Comparative Study on VQ-Based Efficient Mandarin Speech Recognition Method

A VQ based efficient speech recognition method is introduced, and the key parameters of this method are comparatively studied. This method is especially designed for mandarin speaker dependent small size word set recognition. It has less complexity, less resource consumption but higher ARR (accurate recognition rate) compared with traditional HMM or NN approach. A large scale test on the task of 11 mandarin digits recognition shows that the WER(word error rate) can reach 3 86%. This method is suitable for being embedded in PDA (personal digital assistant), mobile phone and so on to perform voice controlling like digits dialing, name dialing, calculating, voice commanding, etc.

Application of VQ-HMM to Chinese Spoken Digit Recognition

In this paper, a new speech recognition method was proposed, which integrated a VQ distortion measure and a discrete HMM. The VQ HMM uses a VQ distortion measure at each state instead of a discrete output probability used by a discrete HMM. The VQ HMM is described, and its speech recognition performance is compared with the conventional HMMs through the experiments on speaker independent Chinese spoken digit recognition. The comparisons confirm that the new method over performed traditional HMMs.

RECOGNITION OF SPOKEN CHINESE DIGIT BASED ON INTEGRATION OF VQ AND HMM

A new speech recognition method is proposed, that integrates a VQ distortion measure and a discrete HMM. This VQ distortion based HMM uses a VQ distortion measure at each state instead of a discrete probability output used by a discrete HMM. Although this method is regarded as a refined version of the VQ distortion based recognition method proposed by Burton et al, it is also considered as a special case of a mixed distribution density HMM. In this paper, the VQ distortion based HMM is described, and it is compared with the conventional HMMs and their speech recognition performance through the experiments on speaker independent spoken digit recognition. From these comparisons, we confirm that the new method is better than the traditional HMMs.

Handwritten digit recognition based on ghost imaging with deep learning

We present a ghost handwritten digit recognition method for the unknown handwritten digits based on ghost imaging(GI)with deep neural network,where a few detection signals from the bucket detector,generated by the cosine transform speckle,are used as the characteristic information and the input of the designed deep neural network(DNN),and the output of the DNN is the classification.The results show that the proposed scheme has a higher recognition accuracy(as high as 98%for the simulations,and 91%for the experiments)with a smaller sampling ratio(say 12.76%).With the increase of the sampling ratio,the recognition accuracy is enhanced.Compared with the traditional recognition scheme using the same DNN structure,the proposed scheme has slightly better performance with a lower complexity and non-locality property.The proposed scheme provides a promising way for remote sensing.

Constructing an AI Compiler for ARM Cortex-M Devices

The diversity of software and hardware forces programmers to spend a great deal of time optimizing their source code,which often requires specific treatment for each platform.The problem becomes critical on embedded devices,where computational and memory resources are strictly constrained.Compilers play an essential role in deploying source code on a target device through the backend.In this work,a novel backend for the Open Neural Network Compiler(ONNC)is proposed,which exploits machine learning to optimize code for the ARM Cortex-M device.The backend requires minimal changes to Open Neural Network Exchange(ONNX)models.Several novel optimization techniques are also incorporated in the backend,such as quantizing the ONNX model’s weight and automatically tuning the dimensions of operators in computations.The performance of the proposed framework is evaluated for two applications:handwritten digit recognition on the Modified National Institute of Standards and Technology(MNIST)dataset and model,and image classification on the Canadian Institute For Advanced Research and 10(CIFAR-10)dataset with the AlexNet-Light model.The system achieves 98.90%and 90.55%accuracy for handwritten digit recognition and image classification,respectively.Furthermore,the proposed architecture is significantly more lightweight than other state-of-theart models in terms of both computation time and generated source code complexity.From the system perspective,this work provides a novel approach to deploying direct computations from the available ONNX models to target devices by optimizing compilers while maintaining high efficiency in accuracy performance.

Switching plasticity in compensated ferrimagnetic multilayers for neuromorphic computing

Current-induced multilevel magnetization switching in ferrimagnetic spintronic devices is highly pursued for the application in neuromorphic computing.In this work,we demonstrate the switching plasticity in Co/Gd ferrimagnetic multilayers where the binary states magnetization switching induced by spin–orbit toque can be tuned into a multistate one as decreasing the domain nucleation barrier.Therefore,the switching plasticity can be tuned by the perpendicular magnetic anisotropy of the multilayers and the in-plane magnetic field.Moreover,we used the switching plasticity of Co/Gd multilayers for demonstrating spike timing-dependent plasticity and sigmoid-like activation behavior.This work gives useful guidance to design multilevel spintronic devices which could be applied in high-performance neuromorphic computing.

Kernel principal component analysis network for image classification

In order to classify nonlinear features with a linear classifier and improve the classification accuracy, a deep learning network named kernel principal component analysis network（ KPCANet） is proposed. First, the data is mapped into a higher-dimensional space with kernel principal component analysis to make the data linearly separable. Then a two-layer KPCANet is built to obtain the principal components of the image. Finally, the principal components are classified with a linear classifier. Experimental results showthat the proposed KPCANet is effective in face recognition, object recognition and handwritten digit recognition. It also outperforms principal component analysis network（ PCANet） generally. Besides, KPCANet is invariant to illumination and stable to occlusion and slight deformation.

AN ADAPTIVELY TRAINED KERNEL-BASED NONLINEAR REPRESENTOR FOR HANDWRITTEN DIGIT CLASSIFICATION

In practice, retraining a trained classifier is necessary when novel data become available. This paper adopts an incremental learning procedure to adaptively train a Kernel-based Nonlinear Representor (KNR), a recently presented nonlinear classifier for optimal pattern representation, so that its generalization ability may be evaluated in time-variant situation and a sparser representation is obtained for computationally intensive tasks. The addressed techniques are applied to handwritten digit classification to illustrate the feasibility for pattern recognition.

OBLIQUE PROJECTION REALIZATION OF A KERNEL-BASED NONLINEAR DISCRIMINATOR

Previously, a novel classifier called Kernel-based Nonlinear Discriminator (KND) was proposed to discriminate a pattern class from other classes by minimizing mean effect of the latter. To consider the effect of the target class, this paper introduces an oblique projection algorithm to determine the coefficients of a KND so that it is extended to a new version called extended KND (eKND). In eKND construction, the desired output vector of the target class is obliquely projected onto the relevant subspace along the subspace related to other classes. In addition, a simple technique is proposed to calculate the associated oblique projection operator. Experimental results on handwritten digit recognition show that the algorithm performes better than a KND classifier and some other commonly used classifiers.

Kernel Factor Analysis Algorithm with Varimax

Kernal factor analysis （KFA） with vafimax was proposed by using Mercer kernel function which can map the data in the original space to a high-dimensional feature space, and was compared with the kernel principle component analysis （KPCA）. The results show that the best error rate in handwritten digit recognition by kernel factor analysis with vadmax （4.2%） was superior to KPCA （4.4%）. The KFA with varimax could more accurately image handwritten digit recognition.

Digital modulation recognition based on instantaneous information

A new digital modulation recognition algorithm based on the instantaneous information is proposed to improve the recognition success rate in the low signal noise ratio （SNR）. First denoising of the instantaneous information is optimized by wavelet filter, which can improve the recognition ability at low SNR. Besides the existing 3 key feature parameters, 3 new key feature parameters are proposed to be used as the decision criteria for identifying different types of digital modulation, which simplifies the recognition process and improves the recognition ability at low SNR. The simulations demonstrate that all modulation types of interest have been classified with success rate of no lower than 99 % when SNR is 10dB. Even if the SNR is lower than 5 dB, the success rate is over 95.4% for most of the modulation types.

Part-based methods for handwritten digit recognition

In this paper, we intensively study the behavior of three part-based methods for handwritten digit recognition. The principle of the proposed methods is to represent a handwritten digit image as a set of parts and recognize the image by aggregating the recognition results of individual parts. Since part-based methods do not rely on the global structure of a character, they are expected to be more robust against various delormations which may damage the global structure. The proposed three methods are based on the same principle but different in their details, for example, the way of aggregating the individual results. Thus, those methods have different performances. Experimental results show that even the simplest part-based method can achieve recognition rate as high as 98.42% while the improved one achieved 99.15%, which is comparable or even higher than some state-of-the-art method. This result is important because it reveals that characters can be recognized without their global structure. The results also show that the part-based method has robustness against deformations which usually appear in handwriting.