Noising Processing And Recognition Under The Car Noise Background Isolated Word Speech Signals

The research on speech recognition technology is a hot topic in today’s era. Speech recognition systems such as a car voice recognition systems have been widely used in people’s life.They make people’s life become convenient.People are eager to achieve the aspiration that we can communicate with the machines by voice.Speech recognition system is divided into three parts:front-end processing, pattern matching and recognition. This paper commence the constitutes of speech recognition system, first, give a general introduction of the signals’pretreatment processing. Use the traditional Hamming window function to fram the signals and describe the most commonly used extraction process of Mel Frequency Cepatral Coefficients. Then introduce the four commonly used models in model-matching techniques,focuse on the basic principles and structure of the Dynamic Time Warping and Hidden Markov Model.Give a detailed introduction of the HMM’s three problems and the the solutions of them.Then improve the drawbacks of the traditional dual-threshold endpoint detection method under the vehicle noise. There are five ways in anti-noise speech recognition technology, the paper make a correction on the power spectral subtraction. Finally, we make a sampling under the specific vehicle noise in laboratory and give a satisfactory result by the simulation of matlab.It lay a theoretical foundation for the vehicle speech recognition system.