Application On Speech Enhancement Method In Vehicle Environment

Due to the economy rapidly development, automobile have undertaken an important role in our modern urban life. Vehicle operational function is getting numerous attentions as drivers are getting inconvenient manipulating the electoral device during driving. In such cases, speech control is the most intensive function in automobile.The quality of voice in car environment are stirring by complex conditions such as noises of engine shaking, friction between car frameworks, friction between air and car, the speech of other passengers are effecting the performance of speech control system. Human can identify these conditions to determine which speech is worth to take, but speech control and identification system is not as ingenious as human ears. Therefore, the research of vehicle environment has unprecedented value.In the first place, this paper introduces the basic knowledge and primary speech enhancement model and five classic algorithms. Speech enhancement in vehicle condition intend to restrain noise from original resource and magnify the object speech. Base on the research from domestic and foreign enhancement method, microphone array is the most effective method till now. It is can fulfill the insufficient cases of single microphone system.On the basis of the fundamental principle analyzes two voice enhancement methods based on vehicle acoustics environment, analyzing shows that one part of the voice contains background noise only such as engine shaking, friction between air and car. The other part of voice contains more than noise also contains speech with expected value. Analysis also proved that the energy of the noise in vehicle cases will concentrated in low-band of the spectrum but the energy of expected voice will locate at high-band in spectrum. This significant trait of speech circumstance in vehicle could make it possible to restrain noise and expend the expected value. In order to improve normal LCMV speech enhancement method, this paper provides an adaptive learning rate beam-form method. In the first place, according the analysis, this method will determine the predictive residual between two channels. At last, the learning rate is adaptively allocated to the noise part, high frequency of speech and low frequency of speech paragraph. The mount of the learning rate will increase depends on the probability of the noise.The result of experiment make it more active than LCMV in vehicle condition.