The Research On Speech Signal Dereverberation Technology

The reverberation of recorded speech signal will be more obvious when the distance of the audio source is far away from the microphone in relatively enclosed space, such as conference room、concert hall etc. Strong reverberation makes pinches of speech overlap each other, weakened speech clarity, and furthermore results the speech confusing, difficult to distinguish, seriously affecting the rate of automatic speech recognition system. Speech dereverberation is an important part in technology of speech enhancement. Simultaneously, it provides pre-processing of speech signal processing such as speech source localization and speech recognition. Speech dereverberation technology is also widely used in many acoustic fields such as ocean acoustics, ultrasonic measure, seismic data analysis, radar and sonar, architecture acoustics. Specific works of the thesis include:First, the principle of reverberation and its characteristics, reverberation time and mathematic model are introduced in this thesis. And elaborated to evaluation methods of speech dereverberation, which includes three subjective evaluation methods and two objective evaluation methods. objective evaluation method include a time domain evaluation and two frequency domain dereverberation indicators. Due to the limitions of gathering indoor reverberation speech signal, this thesis introduces three digital reverberation generators simulate the real reverberation environment.Second, the definition and properties of complex cepstrum are discussed. In the signal channel, the complex cepstrum filtering method and minimum phase decomposition method are introduced, simulation and analysis the two methods.Finally, for single microphone dereverberation techniques available information is limited, it is difficult to obtain a better dereverberation effect. This thesis studies the microphone array dereverberation technology, combined beamforming technology with other dereverberation technology. First,DSB speech dereverberation method based on fixed beamforming and TF-GSC speech dereverberation method based on adaptive beamforming these two beamforming technologies are described. Simulation and analysis DSB and TF-GSC speech dereverberation methods. Because of their good inhibitory effect on the beam direction reverberation, combined DSB with the phase decomposition and due to the problem of this method have the amount of calculation. In this paper, improved dereverberation method based on DSB and minimum phase decomposition algorithm is researched, this method uses phase of reverberation speech signal received by a microphone instead of the all-pass component of the phase information to reduce the amount of calculation. Finally, combined TF-GSC method with post position Wiener filtering technology, we obtained a better robustness speech dereverberation method. After comparing the simulation results, Dereverberation methods combined beamforming technologies with others dereverberation methods have good dereverberation results.