On the performance of isolated word speech recognizers using vector quantization and temporal energy contours

A new vector quantization technique, called storage- and entropy-constrained multi-stage vector quantization (SECMSVQ) technique, is presented for progressive image transmission (PIT) applications. In the SECMSVQ technique, the rate constraint at each stage of the vector quantizer (VQ), and the storage size of the whole VQ are specified independently before the design. A dynamic programming technique is then used to allocate the storage size to each stage of the VQ optimally so that the quantization distortion of the VQ is minimized. The SECMSVQ is well-suited to constructing a PIT system where the transmission time and storage size are the major concerns. Moreover, simulation results show that the SECMSVQ technique can be used effectively to implement signal compression systems under the storage and rate constraints.

This paper describes the implementation of a speaker identification system with reference to Assamese language. The database consists of speech samples that were collected from 15 (fifteen) speakers for ten Assamese words representing the Assamese digits from 0 (shounyo) to 9 (no). Mel Frequency Cepstral Coefficients (MFCC) are used as features for this study. Two independent speaker identification systems have been built in this paper using Vector Quantization (VQ) and I-vector technique. The system built using the I-vector technique obtains comparatively better identification accuracy for speakers when compared with the system developed using VQ technique. Three different systems have been built for both the techniques based on variable feature size. A maximum accuracy of 92.38% is achieved using I-vector technique with 39 MFCC features.

Face recognition has been widely used to identify faces according to the patterns of mouth, lip, eyes and so on. Existing face recognition techniques provide a reasonable accuracy. However, their effectiveness significantly degrades in presence of uncontrolled conditions such as variations of background light and image sizes. To deal with this problem, we propose a novel approach based on Classified Vector Quantization (CVQ) technique for face recognition. The new approach divides images into some blocks and each block is classified into several patterns. The Vector Quantization technique (VQ) is applied on the vectors of each pattern. A major benefit of this technique is to consider the geometric structure of face and skin correlations which provide recognition based on changes on human's age. In order to evaluate our approach, experiments on some standard image databases have been conducted. The results demonstrate the high recognition rate and accuracy of our technique.

Embedded wavelet encoders possess a number of interesting features for video encoding applications. Among those, we can mention-the ability to encode a picture with precise control over the bit-rate. This differs significantly from the DCT-based methods adopted in the MPEG-4 standard, in which the rate cannot be directly set, being controlled by the quantizer step size instead. We investigate, rate-control strategies for embedded wavelet encoders in an MPEG-4 framework, that take advantage of the precise control that can be obtained for the bit-rate in such coders. We also investigate the use of successive approximation vector quantization in them for replacing the traditionally used successive approximation scalar quantization. The results are encouraging showing that both the use of vector quantization and the adoption of an adequate rate control strategy can improve both objective and subjective quality of video sequences encoded using embedded wavelet encoders.

Vector quantization (VQ) based lossy compression techniques are ideally suited for image archival and distribution applications due to their asymmetrical computational properties. The practical use of VQ is however limited due to complications arising from design and maintenance of codebooks. Model-based VQ (MVQ) is a variant of VQ with the computational properties of VQ but without complications from explicit handling of codebooks. In MVQ, error models combined with a human visual system (HVS) model are used to generate an implicit model codebook. The decoder generates same codebook using the model parameters thus eliminating the need for training algorithm, transmitting and maintaining the codebooks. This algorithm was tested on several NASA images and results were published, M. Manchar et al. (1995). The present authors describe progressive MVQ along the lines of progressive VQ and evaluate its performance on NASA images.

The author reviews major methods of applying the vector quantization (VQ) technique to speech and speaker recognition. These include speech recognition based on the combination of VQ and the DTW/HMM (dynamic time warping/hidden Markov model) technique. VQ-distortion-based recognition, learning VQ algorithms, speaker adaptation by VQ-codebook mapping, and VQ-distortion-based speaker recognition. It is concluded that not only has the VQ technique reduced the amount of computation and storage, but it has also created new ideas for solving various problems in speech/speaker recognition. >

The set partitioning in hierarchical trees (SPIHT) approach for still-image compression proposed by Said and Pearlman (1996) is one of the most efficient embedded monochrome image compression schemes known to date. The algorithm relies on a very efficient scanning and bit-allocation scheme for quantizing the coefficients obtained by a wavelet decomposition of an image. In this paper, we adopt this approach to scan groups (vectors) of wavelet coefficients, and use successive refinement vector quantization (VQ) techniques with staggered bit-allocation to quantize the groups at once. The scheme is named vector SPIHT (VSPIHT). We present discussions on possible models for the distributions of the coefficient vectors, and show how trained classified tree-multistage VQ techniques can be used to efficiently quantize them. Extensive coding results comparing VSPIHT to scalar SPIHT in the mean-squared-error sense, are presented for monochrome images. VSPIHT is found to yield superior performance for most images, especially those with high detail content. The method is also applied to color video coding, where a partially scalable bitstream is generated. We present the coding results on QCIF sequences as compared against H.263.

Enhancements to DTW and VQ decision algorithms for speaker recognition

In this paper, an algorithm for binary codebook design has been used in vector quantization technique, which is used to improve the acceptability of the absolute moment block truncation coding (AMBTC) method. Vector quantization (VQ) method is used to compress the bitmap (the output proposed from the first method (AMBTC)). In this paper, the binary codebook can be engender for many images depending on randomly chosen to the code vectors from a set of binary images vectors, and this codebook is then used to compress all bitmaps of these images. The chosen of the bitmap of image in order to compress it by using this codebook based on the criterion of the average bitmap replacement error (ABPRE). This paper is suitable to reduce bit rates (increase compression ratios) with little reduction of performance (PSNR).

In this paper, a new method of vector quantizer design for image compression using Generic codebook and wavelet transformation is proposed. In the proposed method, Self Organizing Feature Map (SOFM) is used for initial codebook generation. A new scheme of wavelet transformation based Vector Quantization (VQ) technique is proposed to replace the SOFM code vectors by VQ code vectors. The proposed wavelet transform is used to generate wavelet coefficients which are then converted into VQ code vectors. Discrete Cosine Transformation based vector quantization technique is proposed in the existing image compression algorithms with low quality images with greater amount of information loss. Hence to increase the psycho visual quality of the reconstructed image wavelet transformation based vector quantization technique is proposed in this paper. Performance of the proposed work is tested with varying codebook size and various training images. Experimental results show that the reconstructed images obtained by the proposed method are of good quality with better compression ratio and higher Peak Signal–to–Noise Ratio.

The use of vector quantization (VQ) in isolated-word recognition of a 20-word vocabulary is examined. A separate sequence of VQ code books is designed for each word in the recognition vocabulary and input words are classified by performing VQ and finding the sequence of code books that achieve the smallest average distortion. In this paper, critical parameters are noted and the results of parameter studies are presented.

An ideal image compression technique for image data archival, retrieval and distribution would be one with the asymmetrical computational requirements of vector quantization (VQ), but without the complications arising from VQ codebooks. Codebook generation and maintenance are stumbling blocks which have limited the use of VQ as a practical image compression algorithm. Model-based VQ (MVQ), a variant of VQ described here, has the computational properties of VQ but does not require explicit codebooks. The codebooks are internally generated using mean removed error and human visual system (HVS) models. The error model assumed is the Laplacian distribution with mean, (lambda) , computed from a sample of the input image. A Laplacian distribution with mean, (lambda) , is generated with a uniform random number generator. These random numbers are grouped into vectors. These vectors are further conditioned to make them perceptually meaningful by filtering the DCT coefficients from each vector. The DCT coefficients are filtered by multiplying by a weight matrix that is found to be optimal for human perception. The inverse DCT is performed to produced the conditioned vectors for the codebook. The only image dependent parameter used in the generation of codebook is the mean, (lambda) , that is included in the coded file to repeat the codebook generation process for decoding.

Recently, novel vector quantization techniques in multiple nonorthogonal domains for both waveform and model-based signal characterization that give an improved qualitative and quantitative signal coding performance as compared to vector quantization in single domain have been reported. In these techniques, vectors are formed either directly from the signal waveform or from the model parameters extracted from the signal. Then, the vectors are represented in multiple nonorthogonal domains. The encoder chooses the domain that best represents the vector according to a predetermined criterion. An iterative codebook enhancement algorithm, applicable to both waveform and model-based vector quantization in nonorthogonal domains is developed and presented in this paper. In this algorithm, each set of codebooks in a given domain is retrained by the vectors that were best represented by that particular set of codebooks in the most recent iteration. The algorithm is applied successfully and extensive simulation results yield considerable performance enhancement of the vector quantization in nonorthogonal domains, for a given bit rate. Sample results are provided which demonstrate the improved performance.

The author reports on the use of vector quantisation techniques to encode the continuous multivariate distributions modeling the probability of occurrence of an observation within a state of the hidden Markov model (HMM). Standard vector quantisation of the spectral features vectors and a novel vector quantisation approach based on the distribution-free goodness-of-fit methodology are used to obtain codebooks for the representation of the probability distribution functions based on mixtures of Gaussian distributions. Initial speech recognition experiments suggest that vector quantisation techniques can be useful for representing mixtures of Gaussian distributions in HMMs. >

Our earlier algorithm (1996) on neural clustering networks (proved to provide globally optimal cluster solutions) is improved herein for considerably faster convergence. The neural network approach and the sort of clustering (vector quantization) are explained. Then the new method is introduced. Computational results are given which demonstrate that the globally optimal solution may be reliable, obtained by the fast algorithm whose convergence rate is of the same order as that of the K-means clustering algorithm. In specific cases the new algorithm may be even faster than K-means. Latent risks of a poor performance of K-means are visualized and consequences for the potential use of vector quantization are discussed.