Wideband Extension Of Telephone Speech

Nowadays, in most of telecommunication systems, the bandwidth of speech signal is limited to 4kHz.In this case,the limited frequency range leads to restricted speech quality compared to wideband speech. While wideband speech coding technology improves the speech intelligibility and naturalness, telecommunication systems have to be modified. However,it requires a long transitional period to change the telecommunication systems due to economical and historical reasons. One possible solution is to artificially extend the signal bandwidth at the receiving end of telephone system by reconstructing the wideband speech.This process of regenerating wideband spectral from narrowband signals is referred to as bandwidth extension of telephone speech.Bandwidth extension can be divided into training and extending procedures. The purpose of training procedure is to find a proper mapping function by training speech , while the purpose of extending part is to synthesize the speech from the parameter.In this thesis,a bandwidth extension algorithm based on codebook mapping is introduced. This algorithm consists of three main modules: the feature extraction, the extension of spectrum and the extension of excitation.The main tasks in the thesis are as follows:(1) In order to extract feature of speech, LPC-derived cepstral coefficient, LSF-coefficient , Mel-cepstral coefficient are introduced. As narrowband feature parameters , MFCC is compare with LPCC and LPC on the advantages and disadvantages of telephone speech quality.(2) The bandwidth extension algorithms,including codebook mapping,linear mapping and statistic mapping, are studied. A new bandwidth extension algorithm is based on codebook mapping is proposed. It optimizes the mapping structure and improves the search efficiency.(3) The major broadband excitation is analyzed. the synthesized speech excited by white noise is compared with speech excited by spectral folding about spectral distortion. A new spectral folding excitation based on sampling rate conversion is adopted to decrease the distorted bandwidth and increase the quality of synthesized speech.(4) A bandwidth extension system based on codebook mapping is established .We adapt an objective and subjective evaluation criteria to evaluate the spectral distortion from the codebook size and the narrowband feature vector (MFCC/LPCC/LPC).As a objective evaluation method , MOS is proposed to evaluate system performance.