The Algorithm Of Audio Event’s Detection

As an important signal to transmit information, sound signal has been takenresearch scholars’more concern and attention. With further research, peopleunderstand the sound signal more and more in-depth, and use it more and more. Theaccurate detection and identification of sound signal are necessary to carry into thenext processing steps. Sound signal can be divided into the speech signal and thenon-speech signal.For speech signal, the speech signal’s detection is the prerequisiteof speech coding, speaker recognition and other speech processing; for the non-speechsignals, it is a very important significance for people to use them in life to identifywhat the sound signal isaccurately. The detection and identification of non-speechsignals have very good application in determining objects’ quality and safetycontrol,etc.First,this thesis describes the research significance and development process ofaudio event detection.It mainly introducesthe voice and non-voice detection’sdevelopment status and development process.And then it pointed out thatthisthesisreaseach mainly about Voice Activity Detection (VAD) and safetymonitoring abnormal voice recognition system.Then,this thesis describesthe basic principles and algorithms of the audio eventdetection in detail.Because the research on speech signal is more in-depth andcomprehensive, the audio event detection techniques and algorithms can be based onthe framework and technologies of speech signal’s detection. The audio eventdetection system can be divided into two parts, one is characteristic parameterextraction, the other is the pattern matching and model training techniques. For thecharacteristic parameters, they can be divided into three categories.They are the timedomain, frequency domain and homomorphic Cepstral. For pattern matchingalgorithm, it commonly used DTW, HMM, ANN and so on. After introducing theoverall algorithm framework, it respectively introduced VAD’s basic principles andcommonly used classical algorithm; it also describes the abnormal voice recognitionsystem’s development status, technology and system evaluation criteria.And then, this thesisdescribes the new algorithm in detail, it is based on thespectralsubtraction and a new short-logarithmic energy combined VAD algorithm. It mainly introduces spectral subtraction’s principle and role and the newshort-logarithmic energy’s improved thinking and advantages. And it also introducesthe new algorithm’s matlab simulation and analysis of results.Then, this thesis describesthe abnormal voice recognition system in detail. Itmainly introduces a multi-level detection algorithm based on the combination of thetime-domain characteristic parameters and homomorphic cepstral features. In thisthesis the abnormal sound is gunshots. And it aso introduces the C languageimplementation and results’ analysis of the new algorithm.At last,the proposed algorithm are summaried. Some problems and work thatneed more improvements in the future are suggested.