Multi Switched Split Vector Quantization of Narrowband Speech Signals M. Satya Sai Ram, P. Siddaiah, and M. Madhavi Latha Abstract—Vector quantization is a powerful tool for speech coding applications. This paper deals with LPC Coding of speech signals which uses a new technique called Multi Switched Split Vector Quantization (MSSVQ), which is a hybrid of Multi, switched, split vector quantization techniques. The spectral distortion performance, computational complexity, and memory requirements of MSSVQ are compared to split vector quantization (SVQ), multi stage vector quantization(MSVQ) and switched split vector quantization (SSVQ) techniques. It has been proved from results that MSSVQ has better spectral distortion performance, lower computational complexity and lower memory requirements when compared to all the above mentioned product code vector quantization techniques. Computational complexity is measured in floating point operations (flops), and memory requirements is measured in (floats). Keywords—Linear predictive Coding, Multi stage vector quantization, Switched Split vector quantization, Split vector quantization, Line Spectral Frequencies (LSF). I. INTRODUCTION ODING of speech signals is a major impairment in today’s low bit rate tele communication systems. The aim of the quantizers used at the transmission end is to compress the speech signal by reducing the bit rate. Coding of speech signals is a challenging task and has been the focus of intense research effort. This paper deals with Linear predictive coding [1]-[2] (LPC) of narrow band speech signals which uses a novel vector quantization [3] scheme called Multi Switched split vector quantization, that is a hybrid of Multi stage vector quantization [4], Switched Split vector quantization [4], split vector quantization [4]-[5]. In MSSVQ, vector quantizers are formed as a cascade, where the difference between the input vector and quantized vector of one stage is fed as an input to the next successive stages. At each stage the quantized vector is obtained by switching from one codebook to the other connected in parellely. MSSVQ algorithm mainly consists of the following steps a) Selection of a switch b) Extracting the codebook from the trained vectors c) obtaining the quantized vector from a set of M. Satya Sai Ram is with Department of ECE, R.V.R& J.C College of Engineering, Guntur-522019, A.P, India (e-mail: m_satyasairam@yahoo.co.in). P. Siddaiah is with Department of ECE, K. L. College of Engineering, Guntur-522502, A.P, India (e-mail: siddaiah_p@yahoo.com). M. Madhavi Latha is with Department of ECE, J.N.T.U College of Engineering, Hyderabad-500072, A.P, India (e-mail: mlmakkena@yahoo.com). codewords d) Extracting the new trained sequence from the ‘old and quantized’ training sequence e) Repeat steps b to d for the required number of stages. The aim of this article is to provide a general review of MSSVQ, and to compare its performance with other product code vector quantization schemes. The practical limitations, regarding computational complexity and memory requirements as a function of bit rate are discussed. The spectral distortion performance[6] of MSSVQ is evaluated in LSF parameter quantization [7]-[9] for narrow band speech coding. The performance is evaluated by using the spectral distortion method. II. MULTI SWITCHED SPLIT VECTOR QUANTIZATION The basic idea of MSSVQ is to use n stages, m switches and s splits, so as to improve the performance of quantization by decreasing the computational complexity and memory requirements when compared to SVQ, MSVQ, and SSVQ. The use of switch vector quantizer exploits the correlation that exists across all dimensions of the vector space. In each SVQ the 10-dimmensional LSF vector is split into 3 parts of 3, 3, 4 divisions respectively. During codebook generation bits are allocated depending on the frequency of the LSFs. Preference is given to high frequency LSFs, when the number of bits is not divisible by 3. For a particular switch the generation of codebooks at different stages is shown in Fig. 1. Fig. 1 Codebook Generation at different stages C World Academy of Science, Engineering and Technology International Journal of Electrical and Computer Engineering Vol:2, No:1, 2008 35 International Scholarly and Scientific Research & Innovation 2(1) 2008 ISNI:0000000091950263 Open Science Index, Electrical and Computer Engineering Vol:2, No:1, 2008 publications.waset.org/7600/pdf