International Journal of Emerging Technology and Advanced Engineering Website: www.ijetae.com (ISSN 2250-2459, Volume 2, Issue 5, May 2012) 15 Performance Analysis of CS-ACELP Algorithm With variation in Weight Factor for Weighted Speech Analysis Nikunj Tahilramani 1 , Ninad Bhatt 2 1 Research Student, Fr. Conceicao Rodrigues College of Engg., Mumbai, India 2 Research scholar, Veer Narmad South Gujarat university, Surat, Gujarat, India 1 nikpriyu@yahoo.com 2 bhattninad@gmail.com Abstract: G.729 algorithm is working on a analysis by synthesis principal and based on a Conjugate-Structure Algebraic Code Excited Linear Prediction (CS-ACELP) coding model which uses 10 ms speech frames for coder operations. Aim of this research is to observe overall performance of G.729 coder based on variation in the parameter weight actor. In this research, rather than using weight factor adaptively speaker dependent variations in the weight factor are suggested. This in turn produces better weighting speech for fixed weight factor moreover its also reduces complexity of implementation of this algorithm as FLAT algorithm process can now be eliminated from implementation. Here, as weighting speech is a decisive signal for subsequent blocks like open and close loop delay search process ,It ultimately overall speech quality performance of the G.729 coder. This coder is implemented in MATLAB with two different values of weight factor and its effect in terms of recovered speech quality of weighted speech has been investigated and studied. Subjective and Objective analysis on weighted speeches have been carried out in order to observe its overall performance. As can be witnessed from obtained results, with increase in the value of weight factor, overall speech quality reduces. If weight factor is made speaker dependent, it may lead to overall better speech quality and can also reduce the computational load of implemented algorithm because off elimination of FLAT algorithm from computation of weighted speech. Keywords: Weighted speech, speech coding, Conjugate structure-Algebraic Code Excited Linear Prediction. I. INTRODUCTION Coder is operating with a digital signal obtained by first performing telephone bandwidth filtering of the analogue input signal, then sampling it at 8000 Hz, followed by conversion to 16-bit linear PCM for the input to the encoder. The output of the decoder should be converted back to an analogue signal in a same manner. speech quality in different conditions include speech output without any distortion and should be able to withstand with channel errors. Decoder should be able to recover the output speech if frame loss occurs during transmission via channel. The paper is organized as follows: In sect. 2 general description of CS-ACELP speech coder is described. In sect. 3, pre-processing which involves high pass filtering and down scaling is discussed. Sect. 4 describes LP analysis and quantization of CS-ACELP. Sect.5 touches upon Subjective and Objective performance evaluation of proposed coder for different values of weight factors. Finally the concluding remarks are given in sect. 6. II. DESCRIPTION OF CS-ACELP SPEECH CODER Coder is derived from a code-excited linear prediction (CELP) model. At every stage recovered signal is being compared with input signal to that particular stage to minimize the mean square error. The CS-ACELP coder is designed to operate with an sampling rate of 8000 samples per second [15]. The input and output samples are represented using 16-b PCM. The coder works on algorithmic delay of 15ms with 5 ms look-ahead at encoder stage. The encoding principle is shown in Fig. 1. After pre-processing, tenth- order LP analysis is performed and the LP parameters are quantized in the line spectral pair (LSF) domain with 18 bits. for better analysis of parameter like pitch and gain in particular frame with reduced codebook searches, input frame is divided into two subframes of 5 ms each.excitation of each subframe is indicated by contribution of adaptive-codebook and a fixed codebook. These parameters are transmitted and analyzed in each subframe. Excitation of adaptive codebook is denoted by integer delay and fractional delay with fixed resolution. Open loop pitch delay is founded in per frame operation which is based on the perceptually weighted speech signal. Close loop pitch search is done around the open loop pitch delay and the target signal is computed by filtering combination of the LP residual through the weighted synthesis filter.