Research On Speech Quality Objective Measure, Effective Receiving And Error Recovery Technology Algorithms Of Speech Communication

Speech communication is one of the first areas which are studied and come into use. Speech communication technologies based on circuit switching have matured and are used cosmically. But with IP technology springing, speech communications based on IP technology have attracted much attention. IP Speech communication areas include many fields. In data link layer, IP Speech communication relative technologies include low code-rate speech coding, channel coding, channel estimation and equalization, modulation and demodulation, multi-user multiplex. In network layer, they include speech communication protocols, QoS, packet loss recovery technologies over internet.In this thesis, we study on some fields of speech communications, including speech quality objective measure algorithm, blind channel estimation and equalization and error conceal and recovery technology and user multi-user receiving technology. The main contributions of this thesis are as follows:Since the bandwidth of each MFCC filter is not enough to approximate the critical band of human auditory system, a new Mel cepstrum algorithm with improved filter bank is proposed, called Mel cepstrum measure based on human factor cepstral coefficients (hfcc). And because the Mel frequency cepstral coefficients are extracted from the log-compressed energy output from each filter, but not sensitive enough to frequency distortion in speech signals, a method to improve frequency sensitivity of MFCC measure is presented too, to which the key aspect is adding frequency enhanced factor while calculating MFCC. In order to get higher correlation between the assessment measure and frequency distortion, another distance measure is experimented, called frequency band distance (FBD) measure. It does not use MFCC as parameters, but directly use the log-compressed energy output from the N triangle filters as parameters for calculating distance.Based on oversampling the output of an IIR channel, we proposed a special Single-Input-Multiple-Output(SIMO) model. We further show that this SIMO can be departed into a AR model and a FIR SIMO model that has been well studied. Therefore existing estimation algorithms can be applied. We also...