This paper proposes a method of low complexity speech mixing with speech codecs based on predictive coding for multimedia conferences. The proposed method applies a filter state management (FSM) technique to a partial mixing method in order to avoid inconsistency of the filter states of encoders. The inconsistency is created by switching of the encoders when the speakers to be mixed are switched. The results of subjective evaluations of speech quality show that the proposed method avoids the inconsistency, and achieves significantly higher speech quality than the conventional partial mixing method without the FSM and almost the same speech quality as the full mixing method. The complexity evaluation results show that the proposed method achieves much lower complexity than the full mixing method.

A new linear prediction analysis method for multichannel signals was devised, with the goal of enhancing the compression performance of the MPEG-4 Audio Lossless Coding (ALS) compliant encoder and decoder. The multichannel coding tool for this standard carries out an adaptively weighted subtraction of the residual signals of the coding channel from those of the reference channel, both of which are produced by independent linear prediction. Our linear prediction method tries to directly minimize the amplitude of the predicted residual signal after subtraction of the signals of the coding channel, and the method has been implemented in the MPEG-4 ALS codec software. The results of a comprehensive evaluation show that this method reduces the size of a compressed file. The maximum improvement of the compression ratio is 14.6% which is achieved at the cost of a small increase in computational complexity at the encoder and without increase in decoding time. This is a practical method because the compressed bitstream remains compliant with the MPEG-4 ALS standard.

Natural, continuous tone images have a very important property of high correlation of adjacent pixels. Images which we wish to compress are usually non-stationary and can be reasonably modeled as smooth and textured areas separated by edges. This property has been successfully exploited in LOCO-I and CALIC by applying gradient based predictive coding as a major de-correlation tool. However, they only examine the horizontal and vertical gradients, and assume the local edge can only occur in these two directions. Their over-simplified assumptions hurt the robustness of the prediction in higher complex areas. In this paper, we propose an accurate gradient selective prediction (AGSP) algorithm which is designed to perform robustly around any type of image texture. Our method measures local texture information by comparison and selection of normalized scalar representation of the gradients in four directions. An adaptive predictor is formed based on the local gradient information and immediate causal pixels. Local texture properties are also exploited in the context modeling of the prediction error. The results we obtained on a test set of several standard images are encouraging. On the average, our method achieves a compression ratio significantly better than CALIC without noticeably increasing of computational complexity.

To increase the discriminating ability of the speech feature based on linear predictive coding (LPC) and increase its noise robustness, an SNR-dependent arcsin transform is applied to the autocorrelation sequence (ACS) of each analysis frame in a speech signal. Moreover, each component in the ACS is also weighted by the normalized reciprocal of the average magnitude difference function (AMDF) for emphasizing its peak structure. Experimental results for the task of Mandarin digit recognition indicate that the LPC speech feature employing the proposed scheme is more robust than some widely used LPC-based approaches over a wide range of SNR values.

In this paper, we propose an adaptive predictive coding method based on image segmentation for lossless compression. MAR (Multiplicative Autoregressive) predictive coding is an efficient lossless compression scheme. Predictors of the MAR model can be adapted to changes in the local image statistics due to its local image processing. However, the performance of the MAR method is reduced when applied to images whose local statistics change within the block-by-block subdivided image. Furthermore, side-information such as prediction coefficients must be transmitted to the decoder with each block. In order to enhance the compression performance, we improve the MAR coding method by using image segmentation. The proposed MAR predictor can be adapted to the local statistics of the image efficiently at each pixel. Furthermore, less side-information need be transmitted compared with the conventional MAR method.