Suggested Algorithm For Speech Signal Coding

This paper discusses a suggested approach for new kind of Speech Signal Compression Algorithm based on KL transform (Karhunen-Loève transform) which normally used with digital image. Speech signal (which is single sequence) rearranged in form of two dimension square matrix to be suitable for KL transformation. Then to be compressed using different no. of eigen values and eigenvectors. Measuring the performance of applied KL transformation (by evaluating the PSNR and SNR factors) with different numbers of eigen values and vectors studied. High rate of compression with high SNR, PSNR got with so closed speech signal to the original one measured using correlation factor which gives values near to 1. In addition to short time needed for compression operation. Suggested Algorithm For Speech Signal Coding. 132 1Introduction: Speech coding is the application of data compression of digital audio signals containing speech [1]. A variety of techniques have been developed to efficiently represent speech signals in digital form for either transmission or storage [2]. The goal in speech coding is to minimize the distortion at a given bit rate, or minimize the bit rate to reach a given distortion [3]. The recent improvements both in speech coding algorithms and DSP hardware have resulted in an increased use of low rate codes in communication systems.[4] The two most important applications of speech coding are mobile telephony and Voice over IP [1]. 2Speech coding: Speech coding uses speech-specific parameter estimation using audio signal processing techniques to model the speech signal, combined with generic data compression algorithms to represent the resulting modeled parameters in a compact bit stream[5][1]. The central problem in speech coding is to represent the speech signal using as little bits as possible so that quality and intelligibility get damage as little as possible[6]. Speech coding is important in digital mobile phone systems and that’s why speech coding methods have advanced considerably in 10 recent years. Thinking commercially, speech coding is the most important application of speech processing field[5][7] The requirements for a good speech codec (codec = coder-decoder) can be: quality of speech suffers as little as possible the speech is compressed in a small amount of bits coding-decoding yields only small delay codec is not sensitive to errors in transmission of bits coding/decoding is computationally fast the codec should perform well with noisy speech (and if possible with other musical signals etc.) several consequent encodings should not impair the quality too much[7]. There are no perfect codecs satisfying all the requirements because part of the requirements are contradictory. However by making different compromises, a large number of coding standards for different applications have been developed. For instance in the speech codec of a mobile phone all the requirements above are essential, whereas in recording of speech in databases delay, computational load and error resiliency are inessential, only the quality and good compression ratio counts. [7] There are plenty of coding methods but they can be divided in roughly two main classes[7]: waveform coding . source coding (also known as vocoders). In waveform coding an effort is made to retain the waveform of the original signal and the coding is based on quantization and removal of redundancies in the waveform.[8]. Dr. khalil Al saif & Saja Jasem mohammed 133 Waveform encoders typically use Time Domain or Frequency Domain coding and attempt to accurately reproduce the original signal. These general encoders do not assume any previous knowledge about the signal. The decoder output waveform is very similar to the signal input to the coder[9]. In source coding the parameters of speech (the type of excitation, model of vocal tract, formant frequencies...) are coded enabling reconstruction in the decoder.[7]. The advantages of the former algorithms are: simplicity, high quality of reestablished Signal and anti—noise. But its compression ratio is low, it need high transmissibility. This kind of algorithm can not satisfy the requirement when transmissibility is not high[8][10]. Speech Parameter Coding Algorithm can achieve high compression ratio and low bit rate. But there are many shortages, such as: bad quality of reestablished Signal, loss of the nature of the speaker, bad naturality etc. And the parameter coding algorithms is sensitive to the environment noise. So this kind of algorithm can not satisfy the requirement of high quality [5] [8][10], this is shown in fig (1). fig (1): Mean opinion scores for various types of speech coders 3Compression: Compression is used just about everywhere. All the images you get on the web are compressed, most modems use compression, HDTV will be compressed using MPEG-2, and several file systems automatically compress files when stored, and the rest of us do it by hand. We must distinguish between "lossless algorithms", which can reconstruct the original message exactly from the compressed message, and "lossy algorithms", which can only reconstruct an approximation of the original message. Lossless algorithms are typically used for text, and lossy for images and sound[11]. Lossy Compression techniques are compression in which some of the information from the original message sequence is lost. This means the Suggested Algorithm For Speech Signal Coding. 134 original sequences cannot be regenerated from the compressed sequence. Just because information is lost doesn’t mean the quality of the output is reduced. The certain losses in images or sound might be completely imperceptible to a human viewer[11], Transform Coding is one of techniques used in this type of compression. Transform Coding is a way of data encoding that is used in many compression schemes. The data is transformed to an other domain before encoding. The transform should be chosen such that it removes the correlation (or dependence) from the source representation.[12] The idea of transform coding is to transform the input into a different form which can then either be compressed better, or for which we can more easily drop certain terms without as much qualitative loss in the output[11]. In this paper KL transformation was achieved for speech coding/compression. Karhaunen-Loeve Transform, or Principal Component Analysis (PCA) has been a popular technique for many image processing and pattern recognition applications. This transform which is also known as Hotelling Transform is based on the concepts of statistical properties of image pixels or pattern features [13][14]. The KL transformation is also knows as the principal component transformation, the eigenvector transformation or the Hotelling transformation [15]. In 1983, the work of Hemon and Mace was extended by a group of researchers at the University of British Columbia in Canada which culminated in the work of Jones and Levy (1987) [15]. In 1988 Freire and Ulrych applied the KL transformation in a somewhat different manner to the processing of vertical seismic pro_ling data[15]. In Signal processing, signal dependent transforms are those transforms that generate their transform vectors based on either apriori knowledge of the signal or by analysis of the signal [16]. The advantage of such schemes is that they can adjust themselves to the characteristics of the signal. The main disadvantage is that they tend to be more computationally intensive. This is caused by two factors. Firstly, if the transform vectors are not predefined then a fast algorithm for their use cannot be generated. Secondly, there will be a computational overhead for the generation of the vectors in the first place, which can be very large. There is also the problem of storage of the transform vectors. As these vectors are unique for the data set under analysis they must all be stored. This overhead can become relatively large if the number of waveforms in the set is small, or if the length of each waveform is large[16][14]. Dr. khalil Al saif & Saja Jasem mohammed 135 4THE KARHUNEN LOEVE TRANSFORM (KLT): KLT can be used on any signal that comprises a set of correlated data sequences[signal], KLT is the optimal transform in that: • It completely decorrelates the original signal. (the transform coefficients are statistically independent for a Gaussian signal). • It optimizes the repacking of the signal energy, such that most of the signal energy is contained in the fewest transform coefficients. • The total entropy of the signal is minimized. • For any amount of compression the MSE (Mean Square Error) in the reconstruction is minimized. Given these abilities, the KLT should be in widespread use. However, there are several disadvantages to using the KLT, the greatest being the computational overhead required to generate the transform vectors. The transform vectors for the KLT are the eigenvectors of the auto covariance matrix formed from the data set[16]. To generate the KLT vectors the procedure is as follows[16]: 1. Take N sequences each of L data points, X[N][L] . 2. Construct an average signal M