Research On Noise Reduction In Speech Recognition Based On Vehicle Noise

Speech recognition technology is one of the key technologies to realize machine intelligence.It allows that the machine can distinguish and understand the language.Today's speech recognition technology is widely used,However some key technologies in the system still need to be further improved and optimized.This thesis focus on the practicability and effectiveness of speech recognition system under the background of vehicle noise.The basic principles and system components of key technologies,including preprocessing,speech noise reduction and feature extraction are described and the following improvements are made to some of the key technologies in the system.A new method of wavelet packet nodal multi-threshold speech denoising based on the fuzzy control system is proposed.The algorithm uses an improved multiply threshold selection method to replace the traditional threshold selection.A new threshold function is applied to quantize the lowest frequency coefficients after the wavelet packet transform to ensure that the noise can be completely filtered out.The fuzzy controller can make adaptive fuzzy filtering and Optimization for the reconstructed signals and get the final speech.A MFCC feature extraction method is based on wavelet packet neighborhood energy segmentation and weighted average.According to the different distribution characteristics of voiced sound and voiceless in frequency domain,the coefficients are reordered after wavelet packet transformation.The weighted average energy of adjacent wavelet packet coefficients is calculated in the high frequency region and the low frequency region respectively.Finally the MFCC feature extraction technology is combined to extract the feature parameters accurately.The isolated word speech recognition system based on HMM is built in Matlab software.the improved noise reduction algorithm and feature extraction technology are applied to verify the anti-noise and feasibility of the reconstruction algorithm.The results show that the improved noise reduction algorithm can remove the noise signal accurately and effectively improve the signal-to-noise ratio and recognition rate of the speech denoising.The improvement of noise reduction technology lays a practical foundation for the realization of vehicular speech interaction.