We present an end-to-end image compression system based on compressive sensing. The presented system integrates the conventional scheme of compressive sampling (on the entire image) and reconstruction with quantization and entropy coding. The compression performance, in terms of decoded image quality versus data rate, is shown to be comparable with JPEG and significantly better at the low rate range. We study the parameters that influence the system performance, including (i) the choice of sensing matrix, (ii) the trade-off between quantization and compression ratio, and (iii) the reconstruction algorithms. We propose an effective method to select, among all possible combinations of quantization step and compression ratio, the ones that yield the near-best quality at any given bit rate. Furthermore, our proposed image compression system can be directly used in the compressive sensing camera, e.g., the single pixel camera, to construct a hardware compressive sampling system.

This paper describes a method to significantly reduce the complexity of continuous density HMM with only a small degradation in performance. The proposed method is noise-robust and may perform even better than the standard algorithm if training and testing noise conditions are not matched. The method is based on approximating the variance vectors of the Gaussian kernels by a vector quantization (VQ) codebook of a small size. The quantization of the variance vectors is done using an information theoretic distortion measure. Closed form expressions are given for the computation of the VQ codebook and the superiority of the proposed distortion measure over the Euclidean distance is demonstrated. The effectiveness of the proposed method is shown using the connected TI digits database and a noisy version of it. For the connected TI digit database, the proposed method shows that by quantizing the variance to 16 levels we can maintain recognition performance within 1% degradation of the original VR system. In comparison, with Euclidean distortion, a size 256 codebook is needed for a similar error rate.

The mobile communication industry in the United States is undergoing major changes. Auctioning of additional spectrum will lead to more service providers and will significantly increase competition. Service providers are likely to customize the services they offer to differentiate themselves from others. We discuss possible technologies for differentiation of services and the implications of these on the requirements for embedded digital signal processors (DSPs). In the US, supporting at least three widely used air interfaces will be a challenge to the industry. The need for customization in the context of multiple standards will create strong pressure to significantly improve the firmware development environment for DSPs. This also implies evolution to architectures that are more friendly to developers.

We present a connected digit recognition system with low storage and computational complexity which achieves good performance in car noise. Our system uses the TI-DIGITS database with additive car noise for training whole-word digit and background models. A digit accuracy of 96.1% is obtained on a 15-speaker database collected in a car using an open microphone with an average SNR of approximately 2 dB. There is a further error reduction of almost 35% if the top two candidate strings are considered using a traceback based N-best algorithm. The system can be implemented on a currently available fixed-point DSP chip. We show that significant performance improvements are obtained by using two-level cepstral mean subtraction (CMS), gender-dependent models and a decoding grammar constraining the possible lengths of digit strings.

In this paper, we describe new techniques to signicantly reduce computational, storage and memory access requirements of a tied-mixture HMM based speech recognition system.Although continuous mixture HMMs oer improved recognition performance, we show that tied-mixture HMMs may oer signicant advantage in complexity reduction for low-cost implementations.In particular, we consider two tasks: (a) connected digit recognition in car noise; and (b) sub-word modeling for command word recognition in a noisy oce environment.We show that quantization of mixture weights can provide an almost three fold reduction in mixture-weight storage requirements without any signicant loss in recognition performance.Furthermore, we show that by combining mixture-weight quantization with techniques such a s V Q-Assist the computational and memory access requirements can be reduced by almost 60-80% without any degradation in recognition performance.

The main results are presented of an analysis of the convergence and the steady-state behavior of the DLMS (delayed least mean square) algorithm, with the aim of providing useful insight which may be helpful in the design of such filters. The problem is defined, and some basic definitions are presented. Conditions for convergence, convergence rate, and limits are discussed. Remarks are presented. The implications of the results for the design of DLMS adaptive filters are addressed.>

It is often desirable to calculate only a few terms of the SVD expansion of a matrix, corresponding to the largest or smallest singular values. Two algorithms, based on gradient and conjugate gradient search, are proposed for this purpose. SVD is computed term by term in a decreasing or increasing order of singular values. The algorithms are simple to implement and are especially advantageous with large matrices.

It is shown that the autoregressive part of an ARMA(p,q) model can be represented as a linear combination of q higher order AR models, and that the coefficients of this linear combination form an approximation to the moving average part. These results are used to develop efficient algorithms for ARMA model estimation.