László Lois

Efficient distance measure for quantization of LSF and its Karhunen-Loeve transformed parameters

This paper presents a new distance measure that is based on the spectral sensitivity of the line spectrum frequency parameters (LSFs) and its Karhunen-Loeve (KL) transformed coefficients. It is shown that the proposed distance measure achieves better performance of vector quantization (VQ) compared to other methods in the field of LSF coding. In most cases, the percentage of outliers is reduced when using the new one, compared to the best results of using other conventional weighting functions, The use of this distance as the weighting function of the LSF transformed parameters is also suggested.

Testbed of a Novel Media Streaming Architecture for Heterogeneous Wireless Environment

The paper presents the testbed of a novel network architecture which aims at providing streaming media services in a heterogeneous environment where the subscribers are roaming among different wired/wireless access networks including ADSL, WiFi, 2.5G and 3G cellular and WiMAX. The architecture provides seamless connectivity during roaming, with adapting the quality of the delivered media stream to the changes of the network characteristics and to the capabilities of a wide variety of devices. The paper presents general design considerations, focuses on a novel vertical handover mechanism, its implementation and evaluation in the testbed, as well as discusses media processing aspects. The different subsystems are currently being integrated into a single testbed which will provide an open environment for experimentation and will be suitable for technology transfer to network service providers and content providers.

Optimal transformation of LSP parameters using neural network

The intraframe correlation properties of line spectrum pair (LSP) are used to develop an efficient encoding algorithm using the Karhunen-Loeve (KL) transformation. The important nonuniform statistical characteristics of LSP frequencies are investigated. Based upon this nonuniform property the neural network based techniques for generating the transform vectors via system training are studied. Using the principal component analysis (PCA) network to decorrelate LSP coefficients, we show that these new approaches lead to as good or better distortion as compared to other methods for speech analysis-synthesis.