Research On Sound Quality Enhancement Algorithms For Digital Speech Communication

The popularity of digital speech communication,especially based on IP networks,has made people pay more attention to the sound quality problems in the communication process.Based on the analysis of various factors affecting sound quality,a set of sound quality enhancement algorithms on both sides of transmitter and receiver are proposed,which involves real-time dereverberation,speaker frequency response equalization and dynamic range compression algorithm.An improved non-intrusive speech quality assessment algorithm is proposed to predict the quality of the communication node,so as to ensure that the sound quality enhancement algorithms can perform parameter tuning according to the application scenario.First of all,the characteristics of digital speech communication system is introduced,both the influencing factors of communication sound quality and the work needed to improve the sound quality are discussed.The audio solution in the actual project is taken as an example for analysis,which provides theoretical and practical basis for selecting the appropriate sound quality enhancement algorithms and implementation structure.Then,based on the voice communication for IP network,the algorithm modules are explained and analyzed,and the effects between different algorithms are compared to figure out schemes suitable for real-time applications and have a long-term effect.Among them,the de-reverberation algorithm,based on the study of block-based spectral subtraction,is a scheme that combined with echo cancellation to improve the overall sound quality of the transmitter.For the receiving end,the compensation for speaker's frequency response and the dynamic range control algorithm of signal amplitude are studied separately.The problem of system instability in the inverse filtering algorithm is solved by the minimum phase decomposition.The voice activity detection and dynamic range control are combined to design a five-stage gain curve,which realizes the noise reduction,gain control and limiting protection functions of the receiving channel.The model of the algorithm modules are simulated on Matlab,and their performance are analyzed in detail.Subsequently,an improved non-intrusive speech quality evaluation algorithm based on QoS is proposed.The mean and variance of Mel cepstral coefficients and its differential coefficients are used as the characteristics.The mixed Gaussian model is trained by the maximum expectation algorithm to predict the speech quality of communication nodes in realtime.It is intended to be used as a reference for parameter tuning of the sound quality enhancement algorithms.Finally,the proposed speech quality assessment algorithm is compared with the standard scheme ITU-T P.563,and an objective MOS score is performed on the aforementioned sound quality enhancement schemes based on the proposed algorithm.After summarizing the running efficiency and execution effect of the enhancement algorithms,it is concluded that the operation results of the algorithms can be expected and applied to the real-time processing platform.At the end of the paper,the direction of further research is pointed out.