Speech is the most important information carrier of our life, but it is vulnerable to interference by noise. Research speech enhancement, because it can separate the useful speech signal from the noisy signal. Speech recognition technology is widely applied in life, speech recognition technology in quiet environment has matured. However, the clean speech signal is always polluted, leading to the recognition rate declined significantly. Robust speech recognition system technology has become a critical problem for the practical use of speech recognition systems. Theoretically, speech enhancement and improved recognition system in noisy environments recognition rate is consistent. Therefore, combining the speech enhancement system to the front of the speech recognition can effectively improve noise resistance of speech recognition systems.Research of this paper divided into three parts. First, we delve into several traditional speech enhancement algorithms, including spectral subtraction, Wiener filtering and the minimum mean square error algorithm. Based on this theory, we simulated these algorithms. The main purpose of experiments is to compare and analyze the advantages and disadvantages of these algorithms. We evaluate these speech enhancement methods from SNR and speech distortion two aspects. Finally, summed up the more appropriate speech enhancement algorithms in different circumstances.Second, we introduce the basic knowledge of speech recognition, emphatically analyzed several key technologies to achieve speech recognition, including pretreatment, endpoint detection, feature extraction and pattern matching. On this basis, we achieved mainstream DTW algorithm and the HMM algorithm. Compared and analyzed the two algorithms through the recognition rate, then we found that the noise resistance of speech recognition system are generally poor.Finally, we applied speech enhancement algorithm to speech recognition. Achieved a robust speech recognition system. Experimental results show that the speech enhancement for robust speech recognition is valid. By comparing several speech enhancement algorithm results and analysis in application of speech recognition, we obtained the better speech enhancement algorithms in speech recognition applications. Combining the advantages of HMM and DTW algorithm in different noise situations, we design a multi-engine speech recognition system. Designed speech recognition system has better noise immunity and higher recognition rate than the traditional method. |