Font Size: a A A

Research Of The Speech Enhancement Methods In Car Speech Recognition System

Posted on:2013-06-25Degree:MasterType:Thesis
Country:ChinaCandidate:L YaoFull Text:PDF
GTID:2248330374951526Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
In recent years, the application of the theory of speech recognition research was supported by the rapid development of speech recognition technology is more and more attention. Speech recognition is the basis for the establishment of computer auditory system, it makes and between computers using natural voice communication possible. In China, the car has been slowly into the lives of people, gave rise to a variety of convenience. Hyundai Motor is more complete, more and more complex electronic equipment, hands off the steering wheel when driving, operating these electronic devices is a very dangerous thing to solve this, so the man-machine interface for in-car electronic devices equipped with voice the best choice of the problem.This article first briefly describes the basic theory of speech recognition, followed by the noise specific to the automotive environment, he proposed to take the wavelet thresholding method, the amplitude spectrum subtraction, the power spectral subtraction methods such as noise, the final proposed using the traditional spectral subtraction and denoising method combining multi-window spectral estimation, and verified by Matlab.Finalized the car background speech signal recognition algorithm design in Matlab volvo noise (SNR respectively.20db,15db,10db,5db,0db) the dtw principle-based speech recognition model, complete endpoint detection, feature extraction, templates, training, and speech recognition algorithm is designed. Take my proposed denoising method, speech recognition rate, respectively99.7%,99.1%.98.7%.For the car of a particular environment. speech recognition research, vehicle noise may affect the speech signal. through the comparison of multiple methods, end-use spectral subtraction and multi-window to estimate the combination of Denoising by comparing the test verified not only reduces the impact of the "musical noise" than simply using the traditional spectral subtraction to enhance the audio quality of the voice, but also improve the SNR of the denoised speech signal. Followed by the establishment of a dtw algorithm-based speech recognition system for vehicle systems, the establishment of a corpus in order to increase the adaptability of the recognition system to improve the recognition rate, while avoiding the system is overly complex, and developed two sets of reference template for the pronunciation of men and women. Different signal to noise ratio, with the noise signal to take a different approach, obtained the recognition rate of the system is set up. Finally pointed out the shortcomings in this study, and the next job outlook and planning.
Keywords/Search Tags:Speech recognition, vehicle noise, spectral subtraction, multi-windowspectral estimation, Matlab
PDF Full Text Request
Related items