| The main content of this paper is the perfection of speech recognition under strong noise. Speech recognition is changing the human voice signal into corresponding technical texts or order by means of identifying and understanding using computers. However, most speech recognition systems is only suitable for the quiet environment, when they are used in the noise environment.,performance is greatly reduced. Therefore, speech recognition under noise environment is difficult at this stage in the development. Nevertheless, it has a significant practical value in rapid development of the information age.The author practises studying endpoint detection of speech recognition and establishes platform including voice collection, synthesis of noise, feature extraction, and recognition result's getting. On the platform, I do improvement studying, to the following:(1)A new endpoint detection algorithm : The study shows that even in a quiet environment, voice recognition systems to identify more than half of the errors, are from the endpoint detector. Therefore, as the necessary step of speech recognition system ,endpoint detection's importance should not be overlooked, particularly noisy environment,and it's accuracy to a great extent have a direct impact on the follow-up work effectively. Therefore, in this paper, a endpoint detection algorithm based on a linear prediction coefficient (LPC) distance, can be an effective solution of noise environment endpoint detection.(2) Based on a new algorithm for improvement: the LPC distance algorithm can effectively curb noise, but it also has its own shortcomings that is not in the high signal-to-noise ratio conditions for effective endpoint detection. This is precisely the strengths of traditional algorithm, the author consider combining the two parameters of the portfolio together. Experiments show that this method can meet a wide range of environmental noise endpoint detection,so it can be in a better position to curb environmental noise. |
PDF Full Text Request