We present some results on audio restoration obtained with an algorithm that solves the problems of broadband noise filtering, signal parameters tracking, and impulsive noise removal by using the Extended Kalman Filter (EKF) theory. We show that, to achieve maximum performance, it is essential to optimize the EKF implementation. To this purpose, having to cope with the nonstationarity of the audio signal, we use two properly combined EKF filters (forward and backward), and introduce a bootstrapping procedure for model tracking. The careful combination of the proposed techniques and an accurate choice of some critical parameters, allows to improve the performance of the EKF algorithm. The presented procedure is validated by listening tests.

Transmission of encoded video sequences over unreliable networks usually requires the adoption of protection techniques to guarantee good reconstruction quality at the receiver side. Multiple description coding (MDC) strategies add reliability to real-time video applications, where retransmission is not possible and packet losses afflict several frames degrading the overall quality. This paper presents a novel MDC scheme based on the H.264/AVC standard and a multiple description scalar quantizer which encodes the residual information into two descriptions. Finally, objective and visual performances are presented.

The standard H.264/AVC defines an efficient coding architecture both for coding applications where bandwidth or storage capacity is limited (e.g., video telephony or video conferencing over mobile channels and devices) and for applications that require high reconstruction quality and bit rate (e.g., HDTV). Since its main applications concern video communication over time-varying bandwidth channels, the bit rate has to be controlled with scalable algorithms that can be implemented on low resource devices. The paper describes a rate control algorithm that needs reduced memory area and complexity compared to other ones. The number of coded bits for each frame is accurately predicted through the percentage of null quantized transform coefficients, which is related to the quantization step via the energy of the quantized signal. It is possible to design a rate control algorithm based on this model that provides a good compression performance at a low computational cost.

In this paper we present a new way to exploit optimal vector prediction theory for lossless compression of color images. To compress color images, the spectral correlation is usually reduced using a color space transformation (e.g., from RGB colour space to YUV Ricoh colour space, or to YCbCr when small losses are allowed). In this work, we exploit the spectral correlation to develop an optimal vector predictor in order to reduce the entropy of the residual image. To this purpose, we consider a pixel as a vector of the three components R, G, and B, and we predict this vector. As a result, we obtain an improvement of the compression ratio at the cost of an increase in the computational complexity. Some techniques to reduce the computational cost are presented. A comparison is carried out with scalar version of GLICBAWLS and JPLG-LS.

This paper presents the investigation carried out for the MetaVision project in the area of lossless compression for digital cinema. The goal of this investigation was to identify, among the many approaches, the one that is most suited to the MetaVision requirements. An extensive simulation campaign has thus identified JPEG-2000 as the best compromise between efficiency, standardization and lossy-to-lossless performance. This result was hardly predictable, especially because of the peculiar kind of material to be coded.

In this paper, we consider the problem of lossless compression of video by taking into account temporal information. Video lossless compression is an interesting possibility in the line of production and contribution. We propose a compression technique which is based on motion compensation, optimal three-dimensional (3-D) linear prediction and context based Golomb-Rice entropy coding. The proposed technique is compared with 3-D extensions of the JPEG-LS standard for still image compression. A compression gain of about 0.8 bit/pel with respect to static JPEG-LS, applied on a frame-by-frame basis, is achievable at a reasonable computational complexity.

In this paper, we consider the problem of lossless compression of video by taking into account temporal information. Video lossless compression is an interesting possibility in the line of production and distribution. We propose a compression technique which is based on motion compensation, optimal 3D linear prediction and context based Golomb-Rice entropy coding. The proposed technique is compared with 3D extensions of the JPEG-LS standard for still image compression. A compression gain of about 0.8 bit/pel with respect to static JPEG-LS, applied on a frame-by-frame basis, is achievable at a reasonable computational complexity.

In a JPEG decoder, the quantized discrete cosine transform (DCT) coefficient blocks are usually processed using the inverse DCT to reconstruct an approximation of the original image. In the presence of quantization of the transform coefficients, the use of the inverse DCT in the decoder is not the best choice: We propose a procedure to design a family of separable 2D synthesis filters that minimize the reconstruction error power and these filters are used in the decoder instead of the inverse DCT. The appropriate reconstruction filters are selected on the basis of the standard quantization information provided in the JPEG bit stream. We show that the proposed decoding method gives some gain with respect to the standard decoder, making use of only the information provided by the JPEG bit stream.

Transform coding is a technique used worldwide for image coding, and JPEG has become the most common tool for image compression. In a JPEG decoder, the quantized transform coefficient blocks are usually processed using the inverse discrete cosine transform (DCT) in order to reconstruct an approximation of the original image. The direct and inverse DCT pair can be arranged in the form of a perfect reconstruction filter bank, and it can be shown that, in the presence of quantization of the transform coefficients, the perfect reconstruction synthesis is not the best choice. In this paper, we propose a procedure for the design of separable 2-D synthesis filters that minimize the reconstruction error power for transform coders. The procedure is used to design a family of filters which are used in the decoder instead of the inverse DCT. The appropriate reconstruction filters are selected on the basis of the standard quantization information provided in the JPEG bit stream. We show that the proposed decoding method gives some gain with respect to the usual decoder in most cases, Moreover, it only makes use of the standard information provided by a JPEG bit stream.

Subband coding is a popular and well established technique used in visual communications, such as image and video transmission. In the absence of quantization and transmission errors, the analysis and synthesis filters in a subband coding scheme can be designed to obtain perfect reconstruction of the input signal, but this is no longer the optimal solution in the presence of quantization of the subband coefficients. We presuppose the use of a two-dimensional (2-D) separable subband scheme and we address the problem of designing, for a given analysis filter bank and assuming uniform quantization of the subband coefficients, the set of row and column synthesis filters that minimize the mean squared reconstruction error at the output of the subband system. Since the corresponding optimization problem is inherently nonlinear, we propose a suboptimal solution that extends a one-dimensional (l-D) optimal filter design procedure, already presented in the literature, to a 2-D separable synthesis filter bank. The separable 2-D extension is not trivial, since the processing in one direction, e.g., the rows, alters the statistics of the signals for the design of the filters in the other direction, e.g., the columns. To further simplify the filter design, we propose to model the input image as a 2-D separable Markov process plus an additive white component. Several design examples using both synthetic signals and real world images are presented, showing that the filters designed using the proposed technique can give a significant gain with respect to the perfect reconstruction solution, especially when the dither technique is used for quantization. The simulation results also show that the proposed image model can be conveniently used in the synthesis filter design procedure.

The complexity of the MPEG video coding standard makes it a challenging task to provide simple strategies to tune the coder parameters in order to produce consistent quality images for a given bit budget. Pre-quantization and pre-coding of individual frames have been proposed in the literature as a viable solution to the problem. In this paper, we show that a simple statistical model for the DCT coefficients can be used to select quantization parameters at the macroblock level. In particular, 8/spl times/8 blocks of DCT coefficients are modelled as i.i.d. random vectors of independent variables. The rate-distortion performance associated to a given quantizer can be calculated from the model and the quantizer parameters can be locally controlled to meet the specified rate-distortion target. The experimental results show that, with a limited complexity, it is possible to achieve a great precision in the bit rate control without actual pre-coding.

We develop an adaptive scheme for quantization of subband or transform coded frames in a typical video sequence coder. Using a generalized Gaussian model for the subband or transform coefficients, we present a procedure to determine the optimum dead-zone quantizer for a given entropy of the quantizer output symbols. We find that, at low bit rates, the dead-zone quantizer offers better performance than the uniform quantizer. The model is used to develop an adaptive procedure to update the quantizer parameters on the basis of the state of a channel buffer with constant output rate and variable input rate. We compare the accuracy of the generalized Gaussian model in predicting the actual bit rate to that achievable using the simpler and more common Laplacian model. Experimental results show that the generalized Gaussian model has superior performance than the Laplacian model, and that it can be effectively used in a practical scheme for buffer control.

A procedure to evaluate the coding gain for 2-D subband systems is explicitly presented. The technique operates in the signal domain and requires the knowledge of the input process auto-correlation function. Both the case of uniform subband and pyramid decomposition are considered. In the case of a separable input process spectrum, the evaluation can be performed by considering appropriately defined 1-D systems, thus, making the procedure very convenient in terms of computational complexity. Using a model that has been recently derived for difference images in motion-compensated image sequence coders, we compare the performance of several filter banks and transform coders in terms of coding gain and asymptotic rate-distortion figures. The results for intraframe and interframe coding show that uniform subband coders can have a performance superior to that of transform coders. Pyramidal schemes appear to have a slightly worse performance.

Symmetric signals are defined by subsets of their support which depend on the specific symmetry. Depending on the symmetry, the cardinality of these subsets may be much smaller than that of the supports. The DFT of the symmetric signals enjoys symmetries related to the input symmetry, by which the DFT is defined by subsets the same as those defining the input signal. In principle, the computation of the DFT of symmetric signal can only use the input subset sufficient for defining the signal. In practice, all known MD FFT algorithms can only take advantage of even and odd parity's symmetries, and cannot exploit general signal symmetries for reducing the computation. This work presents a new MD FFT algorithm capable of saving computation in front of general signal's symmetries.

SDTV/HDTV conversion has several applications of current interest, such as pyramidal coding for simultaneous SDTV and HDTV output, HDTV display of SDTV signals, and SDTV display of HDTV signals. The effectiveness of multistage filters in this application is established. The design of the filters to use in these configurations is a delicate issue, since it has to balance several, and sometimes conflicting, requirements. The design objectives depend on the specific application, and are ultimately targeted to the visual rendition of the system. These objectives do not fit traditional design specs and call for provisions of visual significance in both the frequency domain and the spatio-temporal domain. This work reports on the findings of some systematic research on this field.>

A bimodal automatic speech recognition system, using simultaneously auditory model and articulatory parameters, is described. Results given for various speaker dependent phonetic recognition experiments, regarding the Italian plosive class, show the usefulness of this approach especially in noisy conditions.>

This work presents an original two-stages MD FFT algorithm where in the first stage the signal is processed by multiplier-free butterflies in such a way that at the second stage the computation only needs 1D FFT's. The proposed method is more efficient than any other MD FFT algorithm known to the authors.>

The interlacing property for the line spectrum frequencies (LSFs) is extended to the LSFs associated to successive order predictor polynomials. The corresponding separation theorem gives the most precise lower and upper bounds of the intervals which the LSFs may belong to.>

Examines the distinctive features of 2D diamond-shaped filters, a class of filters quite important for video interpolation/decimation. The geometrical interpretation of linear programming is proposed as an easy tool for understanding the characteristics of M-D FIR filters minimax optimal and the effects of the constraints most common in video application. The relationships between minimax error, filter orders and bandwidth of diamond-shaped FIR filters are investigated, extending results typical of 1D minimax filter design. The effect of constraining one of the filter orders to a pre-assigned value (provision quite useful if the filter order under consideration determines the number of frame-stores) is studied extensively. In light of the reported characteristics of the diamond-shaped FIR filters their multistage implementation becomes extremely attractive. The advantages and the potential of multi-stage implementation of diamond-shaped filters are discussed.>

Granular camera noise is per se objectionable in high-quality TV and it is a source of bothersome visual artifacts with digitally coded video sequences. The first and second order statistical characteristics of this noise are investigated in this work. The results quantify intuitively expected notions, such as the relative unimportance of chrominance noise with respect to the luminance noise and the spatially colored nature of the noise. Noise reduction techniques can use, as general guidance rules, the indications of the presented analysis.>

An algorithm for LPC parameter optimization in multipulse (MP)-linear-prediction-coding (LPC)-based speech coders is presented. It is shown that by taking the nature of the MP excitation signal into account in the LPC parameter computation, it is possible to improve the effectiveness of the LPC model. This results in a better quality of the reconstructed signal in terms of both objective and subjective criteria. The implementation details of the algorithm are discussed, and experimental results are presented. In particular, a comparison with standard MP LPC techniques is given.>

The work extends multistage implementation of sampling structure conversion to the multidimensional (M-D) case. The issues arising in this task are usefully addressed on the basis of lattice theory. Numerical data supporting the advantages of multistage sampling conversion are presented, and the case of format conversion from the 4/3 to the 16/9 aspect ratio is examined as a study case. The main indication of the present work is that multistage implementation, in the case of systems for sampling structure conversion of video signals, may improve the system characteristics and visual rendition.>

An algorithm for LPC (linear predictive coding) parameter optimization in multipulse (MP)-LPC based speech coders is presented. It is shown that, by taking into account the nature of the MP-excitation signal into LPC parameter computation, it is possible to improve the effectiveness of the LPC model. This results in a better quality of the reconstructed signal in terms both of objective and subjective criteria. The implementation details of the algorithm are discussed and experimental results are presented. In particular a comparison with standard MP-LPC techniques is given.>

Automatic key-section recognition is approached by a two-step procedure. The first step is a screening based on arc/straight-segment classification, which associates to each key section a morphological pattern. The second step measures the string distances by means of nonlinear elastic matching among the key-sections possessing the same morphological description. Arc/straight-segment classification, a typical geometrical form description problem, is studied by two techniques, one based on Euclidean geometry notions and the other based on discrete geometry. The results achieved by the second technique appear superior. The performance of the recognition procedure on key-section databases of practical significance are very satisfactory.>

The authors introduce an algorithm for simultaneous magnitude and group-delay design of FIR (finite impulse response) filters, a multiple criterion optimization problem for which no characterization theory is presently available. The algorithm, derived from differential correction, has global convergence. An experimental comparison with a method for complex polynomial approximation, a problem whose mathematical characterization is known, yields very useful and consistent results. The authors compare the results of two filter design formulations which are conceptually similar but formally different, and they present data potentially useful for the mathematical characterization of simultaneous polynomial approximation in magnitude and group-delay.>

Frequency domain techniques are presented for the evaluation of the autocorrelation in the case of multidimensional sequences. The analysis confirms that the use of the proposed frequency domain techniques leads to significant computation time savings, extremely valuable in practical contexts.>

The authors focus on some peculiarities of 2-D group delay equalization by means of cascades of 2*2 sections, a 2-D equalization technique that has lately received considerable attention. It appears that the group delay performance obtained by just one or two 2*2 section equalizers cannot be significantly improved by using equalizers with a greater number of sections. As a consequence, this type of 2-D group delay equalizer can seldom reach the group delay error performance of its 1-D counterpart. The work shows how the characteristics and the possibilities of 2-D group delay equalization markedly differ from those of the 1-D case.>

The authors propose an extension of the differential correction algorithm for simultaneous design of the real and imaginary part of IIR (infinite impulse response) transfer functions in minimax norm. The method has global convergence, and it can be proved to deliver stable IIR filters. Practical implementations of the algorithm compare very favorably with existing design tools. Actual operation is examined by means of filter design examples.>

A grammatical component to supply information to a text-to-speech system is presented. It is composed of four modules: a lexicon, a morphological recognizer, a syntactic preanalyzer, a parser. The parser is composed of a bottom-up algorithm implementing a context-free grammar for Italian simply as Recursive Transition Networks (RTN). No conditions are introduced on the arcs: this will constitute the topic of future work. The paper deals extensively with Focontours and focus assignment rules in Italian by introducing Pierrehumbert's notational system [4]. We found that an Fopeak usually obtains in coincidence with the focussed syllable: this may be either realized as a fall or as a rise followed or preceded by a trailing/leading Fomovement in the opposite direction. The combination of the two tones High/Low and Low/High can appear also on other constituents - Phonological Words - but the steepness of the jump or dipping is remarkably inferior to the one occurring in focussed constituents.

A variant to a recently proposed autoregressive moving average (ARMA) spectrum estimation technique for time series with gapped data is suggested. It is based on the partial fraction expansion of the power spectrum and exhibits some computational and operational advantages.

A system for the automatic translation of any text of Italian into naturally fluent speech is presented. The system, planned for use in a reading machine for the blind, is build up around a Phonological Processor (hence FP) and synthesizes speech-by joining LPC coded diphones. The FP maps into prosodic structures the phonological rules of Italian. Structural information is provided by such hierarchical prosodic constituents as Syllable (S), Metrical Foot (MF), Phonological Word (PW), Intonationa T Group (IG). Onto these structures, phonological rules are applied such as the "letter-to-sound" rules, automatic word stress rules, internal stress hierarchy rules indicating secondary stress, external sandhi rules, phonological focus assignment rules, logical focus assignment rules.

Linear prediction parameters are critically dependent upon the short-term spectrum of speech and therefore to noises and distorsions introduced by transmission and recording systems. The aim of the present work was to evaluate the performances of an automatic speaker recognition system, based on LPC, on the same speech material recorded in three different conditions: on a quiet room, from dialled up telephones lines via direct hookup and via a suction cup tap. Each of ten speakers spoke an 8s long sentence four times over a two-months period. Sentences were manually segmented and performance evaluation was conducted on phonemes, on breath groups and on the whole sentence using a minimum weighted distance classifier.

Over the past years many different methods for automatic speaker recognition have been successfully proposed and tested. The aim of the paper is to present the results of a comparison carried out on two such approaches, viz. Atal /1/ and Sambur /2/, both using, but in a different manner, the parametric representation of speech derived from the linear prediction model. The experiment was carried out on 500 phrase length utterances of 10 speakers recorded over a three -month period. The results obtained in an identification experiment showed the superiority of the Atal approach, as for recognition rate and sensitivity to intra-speaker variability.

Summary form only given, as follows. The paper deals with the application of linear prediction technique to the speech synthesis of both italian and german languages by Standard Speech Reproducing Units (SSRU), it is by combining elementary speech segments of standardized charac teristics extracted fron utterances of native speakers. The nain feature of the method presented is the possibility of synthesizing in a higly intelligible form any nessage of such languages with a very limited amount of data. So far the use of linear predictive coding of the previously realized SSRU sets allowed a memory occupation less than 16 kb for the synthesis of italian language and less than 32 k-bytes for the combined synthesis of italian and german languages. The data flow rate is about 1 kb/s. A key property of the code with respect to methods previously used (i.e. simple concatenation of original segments ) relies in the possibility of greatly enhancing the naturalness of the synthesized speech by varying pitch, amplitude and duration of the synthetic segments. Further, the quantitative rules obtained for generating the SSRU's are expected to be useful, at least as a preliminary investigation tool, for synthesis-by-rule.