Research On Sound Characteristics Of Environmental Abnormalities

With the rapid development of modern information and network technology, the security of information faced with threat, meanwhile people’s sense of security has been improved. Because the abnormal sounds under the specific environment can reveal the information of accidents and critical affairs effectively, the research on abnormal sound under the specific environment has great significance to the intelligent monitoring and safety protection.In this paper, we investigate and analyze lots of articles and papers about sound recognition. Then based on the investigation, the research on abnormal sound recognition of household environment has been studied as following. Abnormal sound in the household environment includes some non-speech signals, such as sound of the porcelain clash, knock on the door, footsteps, and chair fell to the ground etc. The approaches of feature extractions and recognition algorithm mostly follow the traditional methods of speech signal processing. However, there is distinction between the characteristics of speech signal and non-speech, the traditional methods have obvious defects using at non-speech recognition, we extract the time domain features (Short-term Energy and its first order difference),and frequency domain characteristics, such as Mel frequency Cepstrum Coefficients and its first order difference after analyzing the unique characteristics of abnormal sound. The combination of them has also been applied to household environment abnormal sound recognition system. Gaussian Mixture Model (Gaussian Mixture Model, GMM) is selected as the recognition algorithm.A household environment abnormal sound recognition system which includes time domain and frequency domain analysis module, GMM model training module and recognition module has been set up by Matlab. Feature extraction is contained in both training module and recognition module, short-term Energy and its first-order difference, MFCC and its first-order difference and the combination of them are selected as characteristic parameters.Using the abnormal sound recognition system, a variety of simulation has been made to study the recognition performance of different sound. In order to understand the difference between the abnormal sound and speech, a comparison of identification effects at different frame length and frame shift is listed, and he recognition rate and training time of the system are analyzed under different number of samples, meanwhile, different characteristic parameters and SNR also affect the performance of recognition. In the end, a method of ideal parameter selection has been obtained, this study have a good practical and reference value in the analysis of non-speech signals and abnormal sounds.