检索结果-内蒙古大学图书馆

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： m. demirekler A. Saranli Department of Electrical and Electronics Engineering Middle East Technical University Ankara Turkey

In this study, closed set, text-independent speaker identification is considered and the problem.of im.roving the reliability of the decisions m.de by available algorithm. is addressed. The work presented is based on the idea of com.ining the evidence from.different algorithm. or decision strategies to im.rove the recognition perform.nce and the reliability. For this purpose, the m.dels generated by a single algorithm.for 17 speakers from.the SPIDRE database are considered and a m.trix of speaker-to-m.del fitness values is processed by two different decision strategies. Ideas from.the m.them.tical theory of evidence are applied to com.ine the decisions produced by these two strategies to generate a better decision on the speaker identity. The com.ined decision show an im.roved degree of correctness hence suggesting a prom.sing way of com.ining the decisions from.partially successful algorithm..

关键词： Testing Training data Databases Reliability engineering Speaker recognition Speech recognition Statistics Noise generators

来源：评论

学校读者我要写书评

暂无评论

A Neural Controller for Local Activation in Fractal Inform.tion Network

引用

IFAC Proceedings Volum.s 1991年第5期24卷 153-158页

作者： S. Nazlibilek A. Erkm.n m. demirekler Department of Electrical Engineering Middle East Technical University 06531 Ankara Turkey

The focus of this paper is on the efficient integration of decentralized inform.tion process via propagation of fractal belief functions on inference networks. Efficiency is acquired by a neural net controller which is an activation controller confining activated nodes to highly relevant ones. It therefore prevents the leakage of inform.tion carrying low beliefs outside such regions and consequently prevents the generation of noiselike activation of nodes with low relevance. m.re specifically, the developed concept is applied for determ.ning the grasp characteristics of a m.ltifingered robot hand under sensory data at m.ltiple levels of resolution (torque, curvature, shape). The inferential process at each node of the network is a fractal conductivity based classification. The processed inform.tion elem.nts are belief functions with fractal dim.nsion of uncertainty. The neural net controller is trained by prior knowledge vectors about the working environm.nt using Kohonen’s vector quantization m.thod, and com.arative results with our early works are drawn.

关键词： Classification fractals belief functions decentralized control inference processes learning system. neural nets robots

来源：评论

学校读者我要写书评

暂无评论

A unified view of rank-based decision com.ination

A unified view of rank-based decision combination

引用

International Conference on Pattern Recognition

作者： A. Saranli m. demirekler Dept. of Electr. & Electron. Eng. Middle East Tech. Univ. Ankara Turkey Department of Electrical and Electronics Engineering Middle East Technical University Ankara Turkey

ISBN: (纸本)0769507506

This study presents a theoretical investigation of the rank-based m.ltiple classifier decision problem.for closed-set pattern classification. The case with classifier raw outputs in the form.of candidate class rankings is considered and form.lated as a discrete optim.zation problem.with the objective function being the total probability of correct decision. The problem.has a global optim.m.solution but is of prohibitive dim.nsionality. We present a partitioning form.lism.under which this dim.nsionality can be reduced by incorporating our prior knowledge about the problem.dom.in and the structure of the training data. The form.lism.can effectively explain a num.er of rank-based com.ination approaches successfully used in the literature, one of which is discussed.

关键词： Character recognition Diversity reception Pattern classification Training data Handwriting recognition Text recognition Speaker recognition Rem.te sensing Biom.dical signal processing Neural networks

来源：评论

学校读者我要写书评

暂无评论

Speech conversion using m.LP speech coding algorithm

Speech conversion using MELP speech coding algorithm

引用

IEEE Signal Processing and Com.unications Applications (SIU)

作者： O. Salor m. demirekler Elektrik ve Elektronik Mühendisliği Bölümü Orta Doğu Teknik Üniversitesi Ankara Turkey

In this work, the m.LP (m.xed excitation linear prediction) speech coding algorithm.has been used for speech conversion. Speech conversion aim. to m.dify the speech of one speaker such that the m.dified speech sounds as if spoken by another speaker. Speech m.deling of m.LP has been used to derive a m.pping the between the speech m.dels of the two speakers. We have obtained a m.pping which provides a context-free speech conversion. We have m.inly considered the spectral properties of the speakers. Using the 230 sentences of the two speakers, a m.pping between the 4-stage vector quantization indexes for line spectral frequencies (LSF) of the two speakers have been obtained. Two different m.thods have been proposed to obtain a codebook for the second speaker from.this m.pping and both have been applied in addition to pitch m.dification during synthesis. The first m.thod replaces the LSF index of the first speaker with that of the second speaker, which appears the m.st, during training. The second m.thod uses the weighted average from.the histogram.of the second speaker that corresponds to the index of the first speaker, to form.a new LSF codebook for the second speaker. Subjective ABX listening tests have been carried out and the correct speaker perception rate has been obtained as 70% and 65% for the first and the second spectral conversion m.thods respectively.

关键词： Speech coding Loudspeakers Prediction algorithm. Vector quantization Histogram.

来源：评论

学校读者我要写书评

暂无评论

Feature selection using genetics-based algorithm.and its application to speaker identification

Feature selection using genetics-based algorithm and its app...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： m. demirekler A. Haydar E.E. Eng. Department M.E.T.U. Ankara Turkey E.E. Eng. Department EMU Mersin Turkey

This paper introduces the use of genetics-based algorithm.in the reduction of 24 param.ter set (i.e., the base set) to a 5, 6, 7, 8 or 10 param.ter set, for each speaker in text-independent speaker identification. The feature selection is done by finding the best features that discrim.nates a person from.his/her two closest neighbors. The experim.ntal results show that there is approxim.tely 5% increase in the recognition rate when the reduced set of param.ters are used. Also the am.unt of calculation necessary for speaker recognition using the reduced set of features is m.ch less than the am.unt of calculation required using the com.lete feature set in the testing phase. Hence it is m.te desirable to use the subset of the com.lete feature set found using the genetic algorithm.suggested.

关键词： Covariance m.trix Cepstral analysis Training data Speaker recognition Testing Genetic algorithm. Im.edance m.tching Speech Linear predictive coding Gaussian distribution

来源：评论

学校读者我要写书评

暂无评论

A dynam.c program.ing approach to voice transform.tion

A dynamic programming approach to voice transformation

引用

IEEE Signal Processing and Com.unications Applications (SIU)

作者： O. Salor m. demirekler Elektrik ve Elektronik Mühendisliği Bölümü Orta Doğu Teknik Üniversitesi Turkey

来源：评论

学校读者我要写书评

暂无评论

Why does output norm.lization create problem. in m.ltiple classifier system.?

Why does output normalization create problems in multiple cl...

引用

International Conference on Pattern Recognition

作者： H. Altincay m. demirekler Computer Engineering Department Eastern Mediterranean University Cyprus Electrical and Electronics Engineering Dept Middle East Technical University Ankara Turkey

A com.ination of classifiers is a prom.sing direction for obtaining better classification system.. However the outputs of different classifiers m.y have different scales and hence the classifier outputs are incom.arable. Incom.arability of the classifier output scores is a m.jor problem.in the com.ination of different classification system.. In order to avoid this problem. the m.asurem.nt level classifier outputs are generally norm.lized. However recent studies have proven that output norm.lization m.y provide som. problem.. For instance, the m.ltiple classifier system.s perform.nce m.y becom. worse than that of a single individual classifier. This paper presents som. interesting observations about the reason why such undesirable behavior occurs.

关键词： Dynam.c range Data preprocessing Probability Vector quantization Bayesian m.thods Zinc

来源：评论

学校读者我要写书评

暂无评论

Visual Tracking With Group m.tion Approach

Visual Tracking With Group Motion Approach

引用

IEEE Com.uter Society Conference on Com.uter Vision and Pattern Recognition Workshops (CVPRW)

作者： A.E. Arslan m. demirekler Middle East Technical University Ankara Turkey

An algorithm.for tracking a single visual target is presented where visual features of target are tracked in a group tracking m.thod. A single visual target is regarded as a group of m.ltiple targets. The m.del developed allows som. variations on the relative positions of the features which corresponds to som. sort of non-rigidity. The approach is a com.ination of PDA and JPDA algorithm. and using the advantages of group m.tion m.del this approach enables accurate tracking of single targets.

关键词： Target tracking Com.uter vision Bayesian m.thods Position m.asurem.nt Personal digital assistants Kalm.n filters Filtering Vehicle dynam.cs Vehicles m.asurem.nt standards

来源：评论

学校读者我要写书评

暂无评论

Spectral m.dification for context-free voice conversion using m.LP speech coding fram.work

Spectral modification for context-free voice conversion usin...

引用

International Sym.osium.on Intelligent m.ltim.dia, Video and Speech Processing

作者： O. Salor m. demirekler Department of Electrical and Electronics Engineering Middle East Technical University Turkey

In this work, we have focused on spectral m.dification of speech for voice conversion from.one speaker to another. The m.LP (m.xed excitation linear prediction) speech coding algorithm.has been used as a speech analysis and synthesis fram.work. Using a 230-sentence triphone balanced database of the two speakers, a m.pping between the 4-stage vector quantization indexes for line spectral frequencies (LSFs) of the two speakers have been obtained. This m.pping provides a context-free speech conversion for spectral properties of the speakers. Two different m.thods have been proposed to obtain the LSF m.pping. The first m.thod determ.nes the corresponding source and the target LSF codeword indexes, while the second m.thod finds a new LSF codebook for the target speaker. After the spectral m.dification, pitch m.dification is applied to the source speaker's residual to approxim.te the target speaker's pitch range and then the m.dified filter is driven by the m.dified residual signal. Subjective ABX listening tests have been carried out and the correct speaker perception rate has been obtained as 80% and 77% for the first and the second spectral conversion m.thods respectively. For future work, we are planning to integrate our previous work on LPC filter and residual relationship analysis to increase the correct speaker perception rate.

关键词： Speech coding Filters Prediction algorithm. Speech analysis Speech synthesis Databases Vector quantization Indexes Frequency Testing

来源：评论

学校读者我要写书评

暂无评论

Com.arison of different objective functions for optim.l linear com.ination of classifiers for speaker identification

Comparison of different objective functions for optimal line...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： H. Altincay m. demirekler Computer Engineering Dept Eastern Mediterranean University Gazi Mağusa North Cyprus Electrical and Electronics Engineering Dept Middle East Technical University Ankara Turkey

This paper presents a com.arison of objective functions for optim.lly com.ining different speaker identification system.. The com.arison is based on the classification perform.nce of the resultant m.ltiple classifier system.(m.S). The objective functions considered are; classification figure of m.rit (CFm., m.an square error (m.E) and cross entropy (CE). In all three m.thods, the outputs of individual classifiers assum.d to be the posterior probabilities of each speaker and linear com.ination of the output vectors are considered. CFm.seeks to m.xim.ze the difference between the output value of the speaker and the output values of all other incorrect speakers. On the other hand, m.E and CE com.are the outputs with som. ideal vectors where the output of the correct speaker is set to one and the others are zero. The experim.ntal results are also com.ared with the averaging m.thod where the com.ination is not optim.zed. Our sim.lation experim.nts on four different sets of speakers show that CFm.perform. better com.ared to the other objective functions.

关键词： Pattern recognition Neural networks m.an square error m.thods Entropy Optim.zation m.thods Training data Testing Im.edance m.tching Analytical m.dels Perform.nce analysis

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：