Research And Application On The Algorithm Of Speech Enhancement In Car Noise

As car has become an indispensable part of human life, great changes have taken place in present. With the fast development of science and technology, the basic functions of car can no longer meet the needs of the consumers. People tend to pursuit of automobile intellectualization. Thus, equipping cars with varieties of electronic equipments becomes more and more important, as does the car interaction system. Furthermore, in addition to driving, people need extra complex operations when driving such a car, which will greatly increase the risk of driving. From what has been discussed above, the on-board equipments with a voice control system turns out to be one of the best solutions.However, voice signal is inevitably disturbed by all kinds of noise in a real environment and the voice control system cannot be able to identify the control commands accurately with too much noise. In order to guarantee the stability of voice control system, the studying of the speech enhancement algorithm has practical significance.First of all, this article describes the significance and status quo of speech enhancement algorithm under the vehicle noise, as well as how to evaluate the voice quality after denoising processing. Secondly, it briefly introduces the basic conceptions in voice signal processing, such as speech, noise and properties of human auditory system. Thirdly, we studied some classic speech enhancement algorithms and the car noise according to its effects on speech signal. Finally, this dissertation puts forward two speech enhancement algorithms for automobile noise based on the improvement of traditional minimum mean square error of log-spectral amplitude estimator (MMSE-LSA). The specific research topics are as follows:(1) Puts forward a MMSE-LSA algorithm combined with bidirectional MCRA.Based on traditional MMSE-LSA algorithm, we propose an estimation method for the noncausal prior signal-to-noise ratio (SNR) which is able to accurately estimate the prior SNR. Compared with traditional estimating method which is Decision-Directed, the noncausal estimating algorithm is able to recognize speech endpoint and irregular noise better and reduce the music noise further. As noise estimation has a significant effect on estimating the prior SNR, the bidirectional minima controlled recursive averaging algorithm (MCRA) which is adopted in this article has obvious advantages in terms of tracking accuracy of noise. The simulated results show that the improved algorithm has obvious advantages in performance compared with traditional MMSE-LSA algorithm when processing automobile noise. Even under low SNR, the improved algorithm can still improve signal-to-noise ratio and better restore the pure voice.(2) Puts forward a MMSE-LSA algorithm combined with auditory masking effect.Although the improved MMSE-LSA speech enhancement algorithm can extract the clear speech signal even in low SNR environment, there is certain distortion in some time. This article adopts the principle of human auditory masking threshold to solve this problem. This algorithm estimates short-term amplitude spectrum based on the first algorithm and filters the residual noise through a perceptual weighting filter. The perceptual weighting filter is designed based on the noise power spectrum and the noise masking threshold. The power spectrum is given by bidirectional MCRA algorithm. The simulated experiments show that the algorithm can improve the comfort of enhanced speech based on the advantages of the first algorithm. The novel algorithm can work effectively in car.