| Hardly can any speech processing system avoid interference from various noises. Noise can affect the quality and the intelligibility of speech. What's worse, it may deteriorate the system itself. To reduce the interference, many speech-processors adopt the speech enhancement technique widely. Therefore, it has a meaning of great value to further study this technique.It has two main aims: one is to improve the audio quality, eliminate the background noises, and make the listener feel less tired but more willing to listen, and this is a subjective measurement. The other is to increase the signal-to-noise ratio, and it is an objective measurement. This essay, hoping to achieve some progress in this field in terms of the two aims above, did a lot of study on this technique.Having studied and brought in some recent speech enhancement algorithms, this essay did some tests and improvements as follows:Ⅰ. This essay improved the endpoint detection algorithm based on cepstral distance and suggested a judge mechanism based on four conditions, which will be convenient of voice activity detection.Ⅱ. This essay did some researches about this technique based on short-time spectrum estimation, including spectrum subtraction, wiener filtering and speech enhancement scheme based on minimum mean square error estimation of speech short time spectrum amplitude. Not only the fundamentals of these methods were stated, but also the ways they can be put into practice. Besides, some improvement was made, experiment results were compared, and the connections and differences... |
PDF Full Text Request