Research On Two Kinds Of Abnormal Sound Recognition In Office Environment Based On GMM

Due to the continuous development of science and technology, especially the rapid development of information technology and network, people’s living standards have been greatly improved. At the same time, with the accelerated process of urbanization, information security has become the focus of social concern, whether it is the field of national defense or social security, the role of security monitoring is more and more important.As one of the information carrier, the voice is widely distributed in daily life. Due to the diversity, uncertainty and fuzziness of the sound signal, the research on the sound signal has always been a challenging academic field, especially in the field of sound identification. Sound is an important signal that can represent the environmental condition. Compared with visual information, sound can convey the state of the environment out of sight. In addition, for the protection of personal privacy, and as a branch of intelligent monitoring, audio surveillance can describe and reveal the abnormal situation effectively under the specific environment. Therefore, it is important to study the features of abnormal sound signals in the office environment for intelligent monitoring and security.To solve the problem of monitoring the abnormal sound in the office environment, this thesis makes research on the system from several aspects of hardware design, recognition algorithm, simulation and data analysis. Finally, an ideal parameter configuration scheme of the abnormal sound recognition system based on GMM (Gaussian Mixture Model) in the night office environment is putted forward.This thesis analyzes the functional requirements of the embedded audio acquisition terminal in the office environment. Besides, the design of hardware circuit module and driver for Tiny6410 audio acquisition is studied, which are used to complete the abnormal sound acquisition for the establishment of abnormal sound database of office environment.By using the abnormal sound recognition system in the office environment, which is established by the Matlab and GMM, a lot of simulation has been made based on the research of domestic and foreign sound recognition technology, including time domain and frequency domain analysis of abnormal sound, DTW (Dynamic Time Warping) model training and recognition and GMM model training and recognition. Short-term Energy and its first-order difference, MFCC (Mel-Frequency Cepstral Coefficient) and its first-order difference and the combination of them are selected as feature parameters. Based on the results above, the recognition rate and the training time are compared between different samples, GMM orders, feature parameters, frame length and frame shift through a large number of charts. Then, an ideal parameter selection scheme for office environment audio monitoring system is gained.At the end of the thesis, the thesis summarizes the work it has done, and puts forward the deficiency of the thesis which can be improved. At last, the development of abnormal sound recognition system is prospected.