Research And Application Of Speech Signal Enhancement And Separation In Smart Home

With the development of society, people’s living standard is raising rapidly. Smart home has stepped into people’s daily life gradually. In the model of smart home, it is common to control room appliances using speech signal. Therefore, the speech signal processing is particularly important in the model of smart home. In the paper we focused on the speech signal processing in the front-end of smart home, through which we can get a clear,high signal to noise ratio(SNR)speech by processing the received speech signal. In recent years, speech signal processing has captured more and more attention of people. Some researchers have been working hard on this field and have got great achievements. In a word, speech signal processing is the application of knowledge of digital signal processing on speech signals to achieve the specific needs of people。In the paper, we focused on reducing the noise of speech signal, speech signal separation,speech signal endpoint detection direction and so on. Noise reduction can improve the signal to noise ratio(SNR) of the speech signal and. Traditionally, it used a single microphone to reduce in speech processing. However, in this paper we focued on microphone array technology in noise reduction,including fixed and adaptive beamforming technology. In the environment where there are more than one speakers, the speech signal separation technology can be better applied. In this paper,we proposed a speech separation method which is different from the traditional one. We used Some common algorithm for separating the speech signal, which is called semi-blind source separation. Speech endpoint detection can be used to activate the module of smart home and it can reduce the stand-by power consumption effectively. At the same time,these algorithm has been realized in the module of hardware and DSP6747.The innovations of the paper are as follows: 1、Compared to the traditional microphone array,we have done the judgment of the far field and near field, the algorithm will make a different response to different situations.2、In the paper, EDMA has been used to improve the real-timing of algorithm.3、The traditional method least mean square(LMS) algorithm has been used for semi blind source separation.4、The least squares method has been used for semi blind source separation. 5、The fixed beamforming algorithm and semi blind source separation algorithm have been combined for speech signal separation. 6, Speech endpoint detection algorithm has been used for reducing stand-by power consumption. At the same time, this paper has introduced the theory of each algorithm and verified the feasibility of these algorithm. From the experiment,we can find that these innovations presented in this paper can be well applied in practice.