Robust Speech Recognition In Car Noise Environment

With the prevailing of the car in our daily life, people has claimed the car with morefunctions and could be more safety, so that many car production factories and researchesexplore quicker and more maneuverable control method, no doubt that car electricaldevices with speech recognition technology maybe the best solution for this problem.Speech recognition systems can obtain a very high accuracy in laboratory environment,but the performance will degrade rapidly in noisy environments owing to the mismatchbetween the acoustic models and the testing speech. Therefore, to improve the speechrecognition rate in car noise is one of the hottest researches in car environment. Thepapers mainly focus on the following aspects:(1) Complete a speech recognition system which based on Hidden Markov Model.Set up an integrated speech database and noise speech database according to the differentexperiment, ensure the repeatability of the experiment.(2) The paper proposed the robust feature coefficient which adopted randomprojection in feature space, and certified that the proposed method has the higherrecognition rate and the better performance, improved the robustness of speechrecognition.(3) The paper proposed two-step Winer filter denoise method of front end of speechrecognition, the experiments display that the method enhanced the signal to noise rate andimproved the robustness of the recognition system.