Vector Quantization

A simple yet complex approach to modern sophistication.
In this project we used the MFCC approach to build a unique and accurate coefficients extracting processor to extract feature from the voice stored in the database, then on the next stage we again emoployed the use of MFCC to extract coefficients and used vector quantization to match with the user database. For vector Quantization we used Linde-Buzo Grey Algorithm..
Code is available on demand. Send me a private message if you want it.

This thesis encapsulates a study conducted to use Automatic Speech Recognition (ASR) techniques to identify dog barks from ambient recordings. To achieve this, four different classifiers are developed. The first of which relies on cross-correlation, the second on Vector Quantisation (VQ) and the third on an Artificial Neural Network (ANN). The output of each classifier is fed into a fourth ‘Jury’ classifier which determines the overall identification decision.

Mel Frequency Cepstral Coefficients (MFCCs) are used extract features from each recording. The extraction process is optimised by using 30 mel filters spread from 0 to 14 KHz and 15 cepstral coefficients alongside delta features.

The VQ classifier performed the most inconsistently, with little robustness to its applied parameters. Furthermore, its performance degraded considerably when additive white Gaussian noise was introduced in tests. The ANN displayed consistent and robust results.

Of the 537 analysed instances used to test the classifiers, 493 were correctly identified with 22 false positives and 22 false negatives. A second trial was conducted on unseen data in which 61 of the 73 recordings were correctly identified. In both cases a sensitivity of over 89% and specificity of 77% was achieved.

The study successfully concludes that human ASR techniques apply extremely well, and that the use of combining the output of multiple classifiers bolsters performance.

In the recent years, Internet multimedia applications have become very popular. Valuable multimedia content such as digital images, however, is vulnerable to unauthorized access while in storage and during transmission over a network. Streaming digital images also require high network bandwidth for transmission. For effective image transmission over Internet; therefore, both security and bandwidth issues must be considered. In this paper, we present a novel scheme, which combines Discrete Wavelet Transform (DWT) for image compression and block cipher Data Encryption Standard (DES) for image encryption. The simulation results indicate thaf our proposed method enhances the security for image transmission over the Internet as well as improves the transmission rate.

Compression is the art of representing the information in a compact form rather than in its original or uncompressed form. In other words, using the data compression, the size of a particular file can be reduced. This is very useful when processing, storing or transferring a huge file, which needs lots of resources. If the algorithms used to encrypt work properly, there should be a significant difference between the original file and the compressed file. When the data compression is used in a data transmission application, speed is the primary goal. The speed of the transmission depends on the number of bits sent, the time required for the encoder to generate the coded message, and the time required for the decoder to recover the original ensemble. In a data storage application, the degree of compression is the primary concern. Compression can be classified as either lossy or lossless.
Image compression is a key technology in the transmission and storage of digital images because of vast data associated with them. This research suggests an effective approach for image compression using Stationary Wavelet Transform (SWT) and Vector Quantization which is a Linde Buzo Gray (LBG) vector quantization in order to compressed input images in four phases; namely preprocessing, image transformation, zigzag scan, and lossy/lossless compression. Preprocessing phase takes images as input, so that the proposed approach resize the image in accordance with the measured rate of different sizes to (8 × 8) And then converted from (RGB) to (gray scale). Image transformation phase received the resizable gray scale images and produced transformed images using SWT. Zigzag scan phase takes as an input the transformed images in 2D matrix and produced images in 1D matrix. Finally, in lossy/lossless compression phase takes 1D matrix and apply LBG vector quantization as lossy compression techniques and other lossless compression techniques such as Huffman coding and arithmetic coding. The result of our approach gives the highest possible compression ratio and less time possible than other compression approaches. Our approach is useful in the internet image compression.

Emotion detection is a new research era in health informatics and forensic technology. Besides having some challenges, voice based emotion recognition is getting popular, as the situation where the facial image is not available, the voice is the only way to detect the emotional or psychiatric condition of a person. However, the voice signal is so dynamic even in a short-time frame so that, a voice of the same person can differ within a very subtle period of time. Therefore, in this research basically two key criterion have been considered; firstly, this is clear that there is a necessity to partition the training data according to the emotional stage of each individual speaker. Secondly, rather than using the entire voice signal, short time significant frames can be used, which would be enough to identify the emotional condition of the speaker. In this research, Cepstral Coefficient (CC) has been used as voice feature and a fixed valued k-means clustered method has been used for feature classification. The value of k will depend on the number of emotional situations in human physiology is being an evaluation. Consequently, the value of k does not necessarily consider the volume of experimental dataset. In this experiment, three emotional conditions: happy, angry and sad have been detected from eight female and seven male voice signals. This methodology has increased the emotion detection accuracy rate significantly comparing to some recent works and also reduced the CPU time of cluster formation and matching.

In this paper, automatic speaker recognition system is implemented by combining feature extraction and feature matching technique. Feature extraction method that is implemented by the Mel Frequency Cepstral Coefficients (MFCC). The Vector Quantization (VQ) is done using k-means algorithm that is used for the feature matching technique due to high accuracy but its simplicity. A speaker recognition experiments was performed using a 5talkers (3 male and 2 female) with different conditions of Bangla sentence utterances. So, in this proposal achieved 96% efficiency for text dependent or independent speaker recognition system with 60 -90 seconds duration test voice sample.

This paper introduces a new approach to control and drive the DC motor, using speech recognition. The speech signal can be provided through microphone that is connected to computer. A DC motor connected through microcontroller can be driven in forward or reverse direction at different speeds, as well as it can be stopped by giving speech command. In this paper Mel Frequency Cepstral Co-efficient (MFCC) is used to recognize the user's speech and vector quantization is used to increase speech recognition accuracy. The MFCC and VQ algorithm, for speech recognition have been implemented in MATLAB 7.5 version on Windows Vista platform, and the supporting hardware setup is being implemented.

ECTOR QUANTIZATION (VQ), a new direction in source coding, has recently emerged as a powerful and widely applicable coding technique. It was first applied to analysis/synthesis of speech, and has allowed Linear Predictive Coding (LPC) rates to be dramatically reduced to 800 b/s with very slight reduction in quality, and further compressed to rates as low as 150 b/s while retaining intelligibility [ 1,2]. More recently, the technique has found its way to waveform coding [3-51, where its applicability and effectiveness is less obvious and not widely known. There is currently a great need for a low-complexity speech coder at the rate of 16 kb/s which attains essentially "toll" quality, roughly equivalent to that of standard 64-kb/s log PCM codecs. Adaptive DPCM schemes can attain this quality with low complexity for the proposed 3 2 kb/s CCITT standard, but at 16 kb/s the quality of ADPCM or adaptive delta modulation schemes is inadequate. More powerful methods, such as subband coding or transform coding, are capable of producing acceptable speech quality at 16kb/s but have a much higher implementation complexity.

Multi-user wireless systems with multiple antennas can drastically increase the capac-
ity while maintaining the quality of service requirements. The best performance of these
systems is obtained at the presence of instantaneous channel knowledge. Since uplink-
downlink channel reciprocity does not hold in frequency division duplex and broadband
time division duplex systems, efficient channel quantization becomes important. This
thesis focuses on different quantization techniques in a linearly precoded multi-user wire-
less system.
Our work provides three major contributions. First, we come up with an end-to-end
transceiver design, incorporating precoder, receive combining and feedback policy, that
works well at low feedback overhead. Second, we provide optimal bit allocation across the
gain and shape of a complex vector to reduce the quantization error and investigate its
effect in the multiuser wireless system. Third, we design an adaptive differential quantizer
that reduces feedback overhead by utilizing temporal correlation of the channels in a time
varying scenario.

This paper is an overview of the program package LVQ PAK, which has been developed for convenient and e ective application of Learning Vector Quantization algorithms. Two new features are included: fast con ict-free initial distribution of codebook vectors into the class zones, and the optimized-learning-rate algorithm OLVQ1.

he brain is, by far, the most complex system we can record signals from. It can be considered a spatially extended dynamical system with an amazing ability to self-organize not only in response to stimuli coming from the environment but also during internally triggered events like shift of attention or movement planning. To bring the formation of brain activity patterns under experimental study event-related paradigms are usually employed. Signals are recorded while the same well-controlled stimulus is repeatedly applied to human subjects or they repeatedly perform the identical cognitive task. The signal from one such repetition contains the single-trial (ST) response of the brain. In electroencephalographic (EEG) recordings, this signal reflects the changes in the electric potentials measured on the scull and can be attributed to the concerted action of thousands of neurons. In magnetoencephalographic (MEG) recordings, the signal reflects the corresponding magnetic field changes measured around the skull.

In this paper, we propose an automatic voice disorder classification system using first two formants of vowels. Five types of voice disorder, namely, cyst, GERD, paralysis, polyp and sulcus, are used in the experiments. Spoken Arabic digits from the voice disordered people are recorded for input. First formant and second formant are extracted from the vowels [Fatha] and [Kasra], which are present in Arabic digits. These four features are then used to classify the voice disorder using two types of classification methods: vector quantization (VQ) and neural networks. In the experiments, neural network performs better than VQ. For female and male speakers, the classification rates are 67.86% and 52.5%, respectively, using neural networks. The best classification rate, which is 78.72%, is obtained for female sulcus disorder.

A fundamental goal of image compression is to reduce the bit rate for transmission or data storage while maintaining an acceptable fidelity or image quality. To improve the classical lossless compression of low efficiency, a method of image lossless compression with high efficiency is presented. In this paper, The image compression scheme contains wavelet transformation and vector quantization. Wavelets allow complex information such as music, speech, images and patterns to be decomposed into elementary forms at different positions and scales and subsequently reconstructed with high precision and The vector quantization algorithms reduces the transmission bit rate or storage. After analyzing and implementing Wavelet transform in lossless compression, a new method of combining vector quantization with wavelet transform to compress medical images is discussed. The result from this presentation will give the better quality of the image.

A computationally efficient, high quality, vector quantization scheme based on a parametric probability density function (PDF) is proposed. In this scheme, the observations are modeled as i.i.d realizations of a multivariate Gaussian mixture density. The mixture model parameters are efficiently estimated using the expectation maximization (EM) algorithm. A low complexity quantization scheme using transform coding and bit allocation techniques which allows for easy mapping from observation to quantized value is developed for both fixed rate and variable rate systems. An attractive feature of this method is that source encoding using the resultant codebook involves very few searches and its computational complexity is minimal and independent of the rate of the system. Furthermore, the proposed scheme is bit scalable and can switch seamlessly between a memoryless quantizer and a quantizer with memory. The usefulness of the approach is demonstrated for speech coding where Gaussian mixture models are used to model speech line spectral frequencies. The performance of the memoryless quantizer is 1-3 bits better than conventional quantization schemes.

In order to achieve a quantisation technique that has low implementation complexity and performance arbitrarily close to that of Laplacian source optimal vector quantisation, a model of optimal geometric piecewise uniform cubic lattice vector quantiser will be proposed. It will be shown that the presented design procedure is not only much easier for realisation than the procedure of optimal vector quantiser design, but also gives the signal to quantisation noise ratio which in some cases differs from that of the optimal vector quantisation for 0.05 dB and less. It will also be demonstrated that the optimal geometric piecewise uniform vector quantiser model can be described with the optimal piecewise linear radial compression function. Further, since the optimal geometric piecewise uniform lattice vector quantisation in the case of large enough subregions number has performances identical to those of optimal vector quantisation, the optimal nonlinear radial compression function for Laplacian source will also be derived. Figure 1. Total distortion dependence on granular region width in the case when dimension (n) is 16 and quantiser rate is (a) R ¼ 4 b/ dimension, (b) R ¼ 8 b/dimension. RADIAL COMPRESSION FUNCTION 865

Vector quantization is a compression technique which is used to compress the image data in the spatial domain. Since it is a lossy technique, so maintaining the image quality and the compression ratio is a difficult task. For this, the codebook which stores the image data should be optimally designed. In this paper, the K-means algorithm is used to optimize the codebook. We have generated the codebooks from LBG, KPE, KFCG and Random Selection Methods. It is found that the Mean Square Error reduces for every iteration and after a certain number of iterations it stops reducing because the optimal value is reached. We can say that the codebook is optimized at that point. The results show that the codebook obtained from KFCG Algorithm has a least Mean Square Error. This shows that KFCG codebook is more nearer to the optimal point and when applied with K-means algorithm gives the best optimization in comparison with other algorithms.

This paper describes the scalable coder -G.729.1 -which has been recently standardized by ITU-T for wideband telephony and voice over IP (VoIP) applications. G.729.1 can operate at 12 different bit rates from 32 down to 8 kbit/s with wideband quality starting at 14kbit/s. This coder is a bitstream interoperable extension of ITU-T G.729 based on three embedded stages: narrowband cascaded CELP coding at 8 and 12 kbit/s, time-domain bandwidth extension (TDBWE) at 14 kbit/s, and split-band MDCT coding with spherical vector quantization (VQ) and pre-echo reduction from 16 to 32 kbit/s. Side information -consisting of signal class, phase, and energy -is transmitted at 12, 14 and 16 kbit/s to improve the resilience and recovery of the decoder in case of frame erasures. The quality, delay, and complexity of G.729.1 are summarized based on ITU-T results.

An image compression method combining discrete
wavelet transform (DWT) and vector quantization (VQ) is
presented. First, a three-level DWT is performed on the original image resulting in ten separate subbands (ten codebooks are generated using the Self Organizing Feature Map algorithm, which are then used in Vector Quantization, of the wavelet transformed subband images, i.e. one codebook for one subband). These subbands are then vector quantized. VQ indices are Huffman coded to increase the compression ratio. A novel iterative error correction scheme is proposed to continuously check the image quality after sending the Huffman coded bitstream of the error codebook indices through the channel so as to improve the peak signal to noise ratio (PSNR) of the reconstructed image. Ten error codebooks (each for each subband of the wavelet transformed image) are also generated for the error correction scheme using the difference between the original and the reconstructed images in the wavelet domain. The proposed method shows better image quality in terms of PSNR at the same compression ratio as compared to otherDWT and VQ based image compression techniques found in the literature. The proposed method of image compression is useful for various applications in which high quality (i.e. high precision) are critical (like criminal investigation, medicalimaging, etc).

This paper presents the architecture and VHDL design of a Two Dimensional Discrete Cosine Transform (2D-DCT) with Quantization and zigzag arrangement. This architecture is used as the core and path in JPEG image compression hardware. The 2D-DCT calculation is made using the 2D-DCT Separability property, such that the whole architecture is divided into two 1D-DCT calculations by using a transpose buffer. Architecture for Quantization and zigzag process is also described in this paper. The quantization process is done using division operation. This design aimed to be implemented in Spartan-3E XC3S500 FPGA. The 2D-DCT architecture uses 1891 Slices, 51I/O pins, and 8 multipliers of one Xilinx Spartan-3E XC3S500E FPGA reaches an operating frequency of 101.35 MHz One input block with 8 x 8 elements of 8 bits each is processed in 6604 ns and pipeline latency is 140 clock cycles .

In this paper we introduce a novel fuzzy vector
quantization algorithm that tries to solve certain problems
related to the implementation of fuzzy cluster analysis in vector
quantization. The proposed method employs an objective
function that combines the merits of fuzzy and crisp clustering
in a uniform fashion. The algorithm’s structure encompasses
two basic design strategies. The first one concerns the
transition from fuzzy mode, where each training vector is
assigned to more than one codewords, to crisp mode where
each training vector is assigned to only one codeword. To
accomplish this, we use analytical conditions that are extracted
by the minimization of the aforementioned objective function.
The second one is a specially designed pattern reduction
module that helps to significantly reduce the computational
cost. This module acts upon a training vector as soon as it is
transferred in crisp mode. The resulting vector quantization
scheme is fast and easy to implement. Finally, simulation
experiments show that the method is efficient, while it appears
to be insensitive with respect to the selection of its design
parameters.

Several methods have been proposed for detection and classification of power quality (PQ) disturbances using wavelet, Hilbert transform, Gabor transform, Gabor-Wigner transform, S transform, and Hilbert-Haung transform. This paper presents a new method for detection and classification of single and combined PQ disturbances using a sparse signal decomposition (SSD) on overcomplete hybrid dictionary (OHD) matrix. The method first decomposes a PQ signal into detail and approximation signals using the proposed SSD technique with an OHD matrix containing impulse and sinusoidal elementary waveforms. The output detail signal adequately captures morphological features of transients (impulsive and oscillatory) and waveform distortions (harmonics and notching). Whereas the approximation signal contains PQ features of fundamental, flicker, dc-offset, and short-and long-duration variations (sags, swells, and interruptions). Thus, the required PQ features are extracted from the detail and approximation signals. Then, a hierarchical decision-tree algorithm is used for classification of single and combined PQ disturbances. The proposed method is tested using both synthetic and microgrid simulated PQ disturbances. Results demonstrate the accuracy and robustness of the method in detection and classification of single and combined PQ disturbances under noiseless and noisy conditions. The method can be easily expanded for compressed sensing based PQ monitoring networks.

In this paper, a novel post-filtering method applied after the logSTSA filter is proposed. Since the post-filter is derived from vector quantization of clean speech database, it has an equivalent effect of imposing clean source spectral constraints on the enhanced speech. When combined with the logSTSA filter, the additional filter can noticeably suppress residual artifacts by effectively lowering the residual white noise of decision-directed estimation as well as reducing the musical noise of maximum likelihood estimation. Compared to the logSTSA enhanced speech, the overall enhanced speech is able to raise the PESQ score by nearly half a point.

This paper presents an efficient and fast encoding of still images using feedforward neural network technique for codebook search. The image to be coded is first clustered into a small subset of neighboring images and then the neural network-based encoder is used to find the best matching code sequences in the codebook. This subset is then used as a candidate set and an exhaustive search is then performed within this subset to find an optimal code sequence which minimizes the perceptual error between coded and decoded images. In this work, a generic codebook is developed using non-causal Differential Pulse Coded Modulation (DPCM) with residual mean removal and vector quantization using Linde, Buzo and Gray (LBG) method. The codebook is analyzed to identify a pattern in the codebook. This pattern is used to train a neural network to obtain the approximate index of the pattern in the codebook. Then, an extensive search is done around this approximate position identified by the neural network to obtain the nearest neighbor of the pattern. Since the candidate set is usually much smaller that the whole code book, there is a substantial saving in codebook search time for coding an image as compared to the traditional method using full codebook search by LBG algorithm.

ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=845604

Clustering is needed in various applications such as biometric person authentication, speech coding and recognition, image compression and information retrieval. Hundreds of clustering methods have been proposed for the task in various fields but, surprisingly, there are few extensive studies actually comparing them. An important question is how much the choice of a clustering method matters for the final pattern recognition application. Our goal is to provide a thorough experimental comparison of clustering methods for text-independent speaker verification. We consider parametric Gaussian mixture model (GMM) and non-parametric vector quantization (VQ) model using the best known clustering algorithms including iterative (K-means, random swap, expectation-maximization), hierarchical (pairwise nearest neighbor, split, split-and-merge), evolutionary (genetic algorithm), neural (self-organizing map) and fuzzy (fuzzy C-means) approaches. We study recognition accuracy, processing time, clustering validity, and correlation of clustering quality and recognition accuracy. Experiments from these complementary observations indicate clustering is not a critical task in speaker recognition and the choice of the algorithm should be based on computational complexity and simplicity of the implementation. This is mainly because of three reasons: the data is not clustered, large models are used and only the best algorithms are considered. For low-order models, choice of the algorithm, however, can have a significant effect.

License-plate location in sensor images plays an important role in vehicle identification for automated transport systems (ATS's). This paper presents a novel method based on vector quantization (VQ) to process vehicle images. The proposed method makes it possible to perform superior picture compression for archival purposes and to support effective location at the same time. As compared with classical approaches, VQ encoding can give some hints about the contents of image regions; such additional information can be exploited to boost location performance.

This paper presents a new approach to
vector quantization (VQ) based image compression,
which uses an improved partition-based fuzzy
clustering algorithm. The proposed algorithm
employs a generalized fuzzy c-means clustering
approach employing improved fuzzy partitions
(called GIFP-FCM) that was proposed as a
modification of the classical fuzzy c-means algorithm
with an aim to reward crisp membership degrees.
This clustering approach, when applied to VQ based
image compression, aptly demonstrates that the
transition from fuzzy to crisp mode is more efficient
compared to the known approaches and is also
independent of the choice of the initial codebook
vector. The technique is also fast and easy to
implement, and has rapid convergence. Several
experimental results are presented to demonstrate its
distinct advantage over other commonly used
algorithms for image compression.

Speech recognition is always looked upon as a fascinating field in human computer interaction. It is one of the fundamental steps towards understanding human recognition and their behavior. This paper explicates the theory and implementation of Speech recognition. This is a speaker-dependent real time isolated word recognizer. The major logic used was to first obtain the feature vectors using LPC which was followed by vector quantization. The quantized vectors were then recognized by measuring the Minimum average distortion.

We propose a new feature, namely, pitch-synchronous discrete cosine transform (PS-DCT), for the task of speaker identification. These features are obtained directly from the voiced segments of the speech signal, without any preemphasis or windowing. The feature vectors are vector quantized, to create one separate codebook for each speaker during training. The performance of the PS-DCT features is shown to be good, and hence it can be used to supplement other features for the speaker identification task. Speaker identification is also performed using Mel-frequency cepstral coefficient (MFCC) features and combined with the proposed features to improve its performance. For this pilot study, 30 speakers (14 female and 16 male) have been picked up randomly from the TIMIT database for the speaker identification task. On this data, both the proposed features and MFCC give an identification accuracy of 90% and 96.7% for codebook sizes of 16 and 32, respectively, and the combined features achieve 100% performance. Apart from the speaker identification task, this work also shows the capability of DCT to capture discriminative information from the speech signal with minimal pre-processing.

Mostly transforms are used for speech data compressions which are lossy algorithms. Such algorithms are tolerable for speech data compression since the loss in quality is not perceived by the human ear. However the vector quantization (VQ) has a potential to give more data compression maintaining the same quality. In this paper we propose speech data compression algorithm using vector quantization technique. We have used VQ algorithms LBG, KPE and FCG. The results table shows computational complexity of these three algorithms. Here we have introduced a new performance parameter Average Fractional Change in Speech Sample (AFCSS). Our FCG algorithm gives far better performance considering mean absolute error, AFCSS and complexity as compared to others.

Introduces a new image coding scheme using lattice vector quantization. The proposed method involves two steps: biorthogonal wavelet transform of the image, and lattice vector quantization of wavelet coefficients. In order to obtain a compromise between minimum distortion and bit rate, we must truncate and scale the lattice suitably. To meet this goal, we need to know how many lattice points lie within the truncated area. We investigate the case of Laplacian sources where surfaces of equal probability are spheres for the L1 metric (pyramids) for arbitrary lattices. We give explicit generating functions for the codebook sizes for the most useful lattices like Zn, Dn, Es, ∧16

A computationally efficient, high quality, vector quantization scheme based on a parametric probability density function (PDF) is proposed. In this scheme, the observations are modeled as i.i.d realizations of a multivariate Gaussian mixture density. The mixture model parameters are efficiently estimated using the expectation maximization (EM) algorithm. A low complexity quantization scheme using transform coding and bit allocation techniques which allows for easy mapping from observation to quantized value is developed for both fixed rate and variable rate systems. An attractive feature of this method is that source encoding using the resultant codebook involves very few searches and its computational complexity is minimal and independent of the rate of the system. Furthermore, the proposed scheme is bit scalable and can switch seamlessly between a memoryless quantizer and a quantizer with memory. The usefulness of the approach is demonstrated for speech coding where Gaussian mixture models are used to model speech line spectral frequencies. The performance of the memoryless quantizer is 1-3 bits better than conventional quantization schemes.

The foremost issues with most of the existing selective encryption (SE) schemes of images are vulnerability to cryptographic and application specific attacks, reduction in the compression performance, insubstantial computational savings relative to full encryption, and lack of bit stream compliance. This paper is the first one that proposes effective schemes for joint vector quantization (VQ) based image compression and SE. We introduce an image adaptive VQ index compression algorithm suitable for SE, effectively combining remapping of indices, entropy, predictive, differential, and search order coding. We then present SE through codebook pseudorandom shuffling and block ciphering of the VQ index image bit-planes, index usage map, prediction information tables and full indices. Experimentally, the results demonstrate the improved performance and effectiveness of the proposed schemes.

Speech coders operating at low bit rates necessitate efficient encoding of the linear predictive coding (LPC) coefficients. Line spectral frequencies (LSF) parameters are currently one of the most efficient choices of transmission parameters for the LPC coefficients. In this paper, an optimized trellis coded vector quantization (TCVQ) scheme for encoding the LSF parameters is developed. When the selection of a proper distortion measure is the most important issue in the design and operation of the encoder, an appropriate weighted distance measure has been used during the TCVQ construction process. Using this distance, we will show that the LSF TCVQ encoder performs better than the encoder conceived with the unweighted distance and a reduction of about 1-2 bits/frame is obtained while maintaining the same performance. We further applied the TCVQ encoder system for encoding the LSF parameters of the US federal standard (FS1016) 4.8 kbps code excited linear prediction (CELP) speech coder. At lower bit rates, our objective and subjective evaluation results show that the incorporated LSF TCVQ encoder performs better than the 34 bits/frame LSF scalar quantizer used originally in the FS1016 coder. The subjective tests reveal also that the 27 bits/frame scheme produces equivalent perceptual quality to that when the LSF parameters are unquantized. r

Voice conversion methods have the objective of transforming speech spoken by a particular source speaker, so that it sounds as if spoken by a different target speaker. The majority of voice conversion methods is based on transforming the short-time spectral envelope of the source speaker, based on derived correspondences between the source and target vectors using training speech data from both speakers. These correspondences are usually obtained by segmenting the spectral vectors of one or both speakers into clusters, using soft (GMM-based) or hard (VQ-based) clustering. Here, we propose that voice conversion performance can be improved by taking advantage of the fact that often the relationship between the source and target vectors is one-to-many. In order to illustrate this, we propose that a VQ approach namely constrained vector quantization (CVQ), can be used for voice conversion. Results indicate that indeed such a relationship between the source and target data exists and can be exploited by following a CVQ-based function for voice conversion.

Clustering algorithms aim at modeling fuzzy (i.e., ambiguous) unlabeled patterns efficiently. Our goal is to propose a theoretical framework where the expressive power of clustering systems can be compared on the basis of a meaningful set of common functional features. Part I of this paper reviews the following issues related to clustering approaches found in the literature: relative (probabilistic) and absolute (possibilistic) fuzzy membership functions and their relationships to the Bayes rule, batch and on-line learning, prototype editing schemes, growing and pruning networks, modular network architectures, topologically perfect mapping, ecological nets and neuro-fuzziness. From this discussion an equivalence between the concepts of fuzzy clustering and soft competitive learning in clustering algorithms is proposed as a unifying framework in the comparison of clustering systems. Moreover, a set of functional attributes is selected for use as dictionary entries in the comparison of clustering algorithms, which is the subject of Part II of this paper [1].

The color satellite image compression technique by vector quantization can be improved either by acting directly on the step of constructing the dictionary or by acting on the quantization step of the input vectors. In this paper, an improvement of the second step has been proposed. The k-nearest neighbor algorithm was used on each axis separately. The three classifications, considered as three independent sources of information, are combined in the framework of the evidence theory. The best code vector is then selected, after the image is quantized, Huffman schemes compression is applied for encoding and decoding.

In this paper, we propose an automatic voice disorder classification system using first two formants of vowels. Five types of voice disorder, namely, cyst, GERD, paralysis, polyp and sulcus, are used in the experiments. Spoken Arabic digits from the voice disordered people are recorded for input. First formant and second formant are extracted from the vowels [Fatha] and [Kasra], which

Predictive rate-distortion (RD) optimized motion estimation techniques are studied and developed for very low bit-rate video coding. Four types of predictors are studied: mean, weighted mean, median, and statistical mean. The weighted mean is obtained using conventional linear prediction techniques. The statistical mean is obtained using a finite-state machine modeling method based on dynamic vector quantization. By employing prediction, the motion vector search can then be constrained to a small area. The effective search area is reduced further by varying its size based on the local statistics of the motion field, through using a Lagrangian as the search matching measure and imposing probabilistic models during the search process. The proposed motion estimation techniques are analyzed within a simple DCTbased video coding framework, where an RD criterion is used for alternating among three coding modes for each 8 2 8 block: motion only, motion-compensated prediction and DCT, and intra-DCT. Experimental results indicate that our techniques yield very good computation-performance tradeoffs. When such techniques are applied to an RD optimized H.263 framework at very low bit rates, the resulting H.263 compliant video coder is shown to outperform the H.263 TMN5 coder in terms of compression performance and computations simultaneously.

On one hand, multiple object detection approaches of Hough transform (HT) type and randomized HT type have been extended into an evidence accumulation featured general framework for problem solving, with five key mechanisms elaborated and several extensions of HT and RHT presented. On the other hand, another framework is proposed to integrate typical multi-learner based approaches for problem solving, particularly on Gaussian mixture based data clustering and local subspace learning, multi-sets mixture based object detection and motion estimation, and multi-agent coordinated problem solving. Typical learning algorithms, especially those that base on rival penalized competitive learning (RPCL) and Bayesian Ying-Yang (BYY) learning, are summarized from a unified perspective with new extensions. Furthermore, the two different frameworks are not only examined with one viewed crossly from a perspective of the other, with new insights and extensions, but also further unified into a general problem solving paradigm that consists of five basic mechanisms in terms of acquisition, allocation, amalgamation, admission, and affirmation, or shortly A5 paradigm.

The content based image retrieval (CBIR) technique is one of the most popular and evolving research areas of the digital image processing. The goal of CBIR is to extract visual content like colour, texture or shape, of an image automatically. This paper proposes an image retrieval method that uses colour and texture for feature extraction. This system uses the query by example model. The system allows user to choose the feature on the basis of which retrieval will take place. For the retrieval based on colour feature, RGB and HSV models are taken into consideration. Whereas for texture the GLCM is used for extracting the textural features which then goes into Vector Quantization phase to speed up the retrieval process.

On one hand, multiple object detection approaches of Hough transform (HT) type and randomized HT type have been extended into an evidence accumulation featured general framework for problem solving, with five key mechanisms elaborated and several extensions of HT and RHT presented. On the other hand, another framework is proposed to integrate typical multi-learner based approaches for problem solving, particularly on Gaussian mixture based data clustering and local subspace learning, multi-sets mixture based object detection and motion estimation, and multi-agent coordinated problem solving. Typical learning algorithms, especially those that base on rival penalized competitive learning (RPCL) and Bayesian Ying-Yang (BYY) learning, are summarized from a unified perspective with new extensions. Furthermore, the two different frameworks are not only examined with one viewed crossly from a perspective of the other, with new insights and extensions, but also further unified into a general problem solving paradigm that consists of five basic mechanisms in terms of acquisition, allocation, amalgamation, admission, and affirmation, or shortly A5 paradigm.

In Part I of this paper , an equivalence between the concepts of fuzzy clustering and soft competitive learning in clustering algorithms is proposed on the basis of the existing literature. Moreover, a set of functional attributes is selected for use as dictionary entries in the comparison of clustering algorithms.

This paper presents a method for recognizing isolated spoken Bengali numerals. Noisy audio samples have been considered as input in this study. Mel frequency cepstral coefficients (MFCC) have been used for extraction of feature from the audio samples. Vector quantization is applied to reduce the dimension of the feature vectors and to generate a vector codebook for the numerals. The classification is based on the dynamic time warping (DTW) and a minimum distance classifier based on Euclidean distance measure. Both the speaker dependent and speaker independent situations have been considered for checking accuracy. Results show the limitations of MFCC based standard speech processing approach in speaker independent spoken digit recognition scenario in the presence of noise.

The relation between hard c-means (HCM), fuzzy c-means (FCM), fuzzy learning vector quantization (FLVQ), soft competition scheme (SCS) of and probabilistic Gaussian mixtures (GM) have been pointed out recently by . We extend this relation to their training, showing that learning rules by these models to estimate the cluster centers can be seen as approximations to the expectation-maximization (EM) method as applied to Gaussian mixtures. HCM and unsupervised, LVQ use 1-of-c type competition. In FCM and FLVQ, membership is the ¹2/(m ¹ 1)th power of the distance. In SCS and GM, Gaussian function is used. If the Gaussian membership function is used, the weighted within-groups sum of squared errors used as the fuzzy objective function corresponds to the maximum likelihood estimate in Gaussian mixtures with equal priors and covariances. The fuzzy clustering method named fuzzy c-means alternating optimization procedure (FCM-AO) proposed to optimize the former is then equivalent to batch EM and SCS's update rule is a variant of the online version of EM. The advantages of the probabilistic framework are: (i) we no longer have spurious spread parameters that needs fine tuning as m in fuzzy vector quantization or b in SCS; instead we have a variance term that has a sound interpretation and that can be estimated from the sample; (ii) EM guarantees that the likelihood does not decrease, thus it converges to the nearest local optimum; (iii) EM also allows us to estimate the underlying distance norm and the cluster priors which we could not with the other approaches. We compare Gaussian mixtures trained with EM with LVQ (HCM), SCS and FLVQ on the IRIS dataset and see that it is more accurate due to its being able to take into account the covariance information. We finally note that vector quantization is generally an intermediate step before finding a final output for which supervision may be possible. Thus, instead of an uncoupled approach where an unsupervised method is used first to find the cluster parameters followed by supervised training of the mapping based on the memberships, we advocate a coupled approach where the cluster parameters and mapping are trained supervised in a coupled way. The uncoupled approach ignores the error at the outputs which may not be ideal. ᭧

In this letter, a preliminary study of habituation in self-organizing networks is reported. The habituation model implemented allows us to obtain a faster learning process and better clustering performances. The habituable neuron is a generalization of the typical neuron and can be used in many self-organizing network models. The habituation mechanism is implemented in a SOM and the clustering performances of the network are compared to the conscience learning mechanism that follows roughly the same principle but is less sophisticated.

A full-rank under-determined linear system of equations Ax = b has in general infinitely many possible solutions. In recent years there is a growing interest in the sparsest solution of this equation-the one with the fewest non-zero entries, measured by x 0 . Such solutions find applications in signal and image processing, where the topic is typically referred to as "sparse representation". Considering the columns of A as atoms of a dictionary, it is assumed that a given signal b is a linear composition of few such atoms. Recent work established that if the desired solution x is sparse enough, uniqueness of such a result is guaranteed. Also, pursuit algorithms, approximation solvers for the above problem, are guaranteed to succeed in finding this solution.