A Complex Sound Recognition Method In Noise Environment

Today's society has entered the era of artificial intelligence,speech recognition technology has been quite mature.For the complex voice of real life,because of the complexity and diversity of its sound source,combined with the interference of background noise,the recognition of this field is far from mature enough,and there are still many problems and defects.Therefore,it is of great practical value and theoretical value to study the recognition of complex sound in noise environment.Complex sound is a kind of sound signal that contains a variety of sound types and whose boundaries are difficult to distinguish between these sounds.At present,the detection method of this kind of sound mainly adopts the traditional speech recognition technology.Sound signal pronunciation is relatively fixed and the energy is stable,but the type of complex sound is numerous,the sound mechanism is different,instantaneous energy is also large,and will be interference by environmental noise.So only using traditional speech recognition technology can't be better applied to complex sound recognition.In this paper,the following research work is carried out:(1)Firstly,several time-frequency characteristics commonly used in sound recognition are introduced.By extracting and analyzing the characteristic parameters of complex sound samples,it is proposed that the complex sound is described by the combination of time and frequency characteristics,and a variety of mixed features were compared.(2)In the process of complex sound recognition,a training sample selection algorithm based on clustering annotation is proposed to select the sample representative sets more quickly and accurately,and carried out a comparative experiment of different clustering methods.(3)Finally,a complex sound recognition framework based on Hidden Markov Model(HMM)is proposed.The simulation results show that the mixed characteristics combined by Autocorrelation Function and Mel Frequency Cepstral Coefficient can represent complex sound features.The training sample selection algorithm based on neighbor propagation clustering and the HMM model recognition framework can be used to improve the recognition accuracy and efficiency of complex sound in noise environments.