Research On Speech Enhancement Method In Car Environment

Speech control system in a car gets people’s extensive attention due to the inconvenienceand unsafety caused by hands off the steering wheel when drivers operate electrical devices.The quality of speech is deteriorated in car environment by numerous noises such as enginevibration, air turbulence and friction between the road surface and the tire, and these noisesgreatly affect the performance of the whole system. Therefore, the research on speechenhancement in car environment has theoretical and practical significance.Speech enhancement in car environment should be considered as microphone arrayspeech enhancement technique applied to a specific acoustic circumstance. Based on thereview of domestic and foreign speech enhancement methods in car environment, paperintroduces the basic knowledge of microphone array speech enhancement technology and fourclassic algorithms and analyzes two methods of speech enhancement in car environmentwhich are called a subband hybrid beamforming and probabilistic spectral gain beamformingrespectively through the experiments.By analyzing the characteristics of car acoustic shows that the speech in car environmentcan be split into two parts, one of them contains only background noise and the other containsnot only background noise but also expect speech, and the analysis also shows that the energyof noise in car environment is concentrated in the low-band but the energy of expect speech ismainly in the higher band. For the two characteristics of speech in car environment, paperproposes a adaptive learning rate beamforming method. Firstly, the method detemines thecross correlation of linear predictive residual between two channels. Then the crosscorrelation’s contour is utilized as an input to calculate speech presence probability. Finally,the learning rate is adaptively assigned to noise paragraph, high frequency of speechparagraph and low frequency of speech paragraph according to the probability and frequencythreshold value, In order to control amount of noise reduction of the linearly constraintminimum variance (LCMV) algorithm noise capacity and complete the speech enhancement.the experiment resualt shows that the algorithm of paper suppress the noise more effectivelythan LCMV in car environment.