![]() | Only 14 pages are availabe for public view |
Abstract The speech compression using Wavelet Transform (WT) plays a significant role in speech communication. It yields a compact data representation allowing efficient storage and transmission of information. Orthonormal wavelet transform bases are achievable with basis functions being well-localized in time and frequency. As a result, the wavelet transform decomposes the speech into a set of coefficients with different resolution to different frequency bands. The wavelet transform is covered in depth in the thesis. It deals with the representation and construction forms ofWT, its application, its relations and differences with other representations. The relation between WT and sub band coding is discussed. The construction of WT based on iterated filter banks is reviewed. The best choice of wavelet bases for speech compression among a number of well-known wavelet functions is investigated. The best results are achieved by applying the BiorSplines wavelet. Also an investigation on the proper decomposition level is done. The effect of time delay and the transmission channel errors using WT are tested. This thesis introduces two new techniques for speech compression using Wavelet Transform (WT). The first technique, Zero Wavelet transform (ZWT), eliminates the high frequency coefficients of the wavelet decomposition with energy values below a certain threshold level. The second technique, Average Zero Wavelet Transform (AZWT), in addition to fulfilling the goal of the first technique, it averages the approximate coefficients of the wavelet decomposition. These coefficients are almost constant at higher decomposition levels of the transform. The wavelet coefficients are, then, quantized using Lloyd’s algorithm and coded using the entropy coding technique before being transmitted. At the receiver end, the received signal is decoded and dequantized before being processed. These two new techniques offer 35- 44% data compression. They have high speech quality operating at rate of 2.5-16 kb/s and have a satisfactory delay of 20ms. Key Words: Wavelets, data compression, subband coding, Signal processing, Speech representation. |