Research On Intelligent Audio Detection And Enhancement Method In Strong Noise Background

Audio signal detection and enhancement is the basis of many applications such as speech recognition,music classification,environmental monitoring,abnormal event alarm,etc.,and is of great significance in modern social life.However,the difficulty of audio signal detection and enhancement in a strong noise environment is also increasing.Therefore,improving detection accuracy and reducing noise interference has always been a research hotspot in the field of audio signal detection and enhancement.Aiming at the above difficult problems,this paper conducts research on audio intelligent detection and enhancement methods under strong noise background,which mainly includes the following three parts: 1)Research on audio endpoint detection method,realize the distinction between target audio segment and noise segment in complex noise environment;2)Research on audio type recognition method,realize the type judgment of target audio segment and noise segment through deep learning model;3)Research on audio enhancement method,based on the noise type identification results,select the optimal combination of parameters in the statistical method to achieve audio enhancement.Aiming at the problem of distinguishing the audio segment and the noise segment in the noisy signal,an audio endpoint detection method based on naive Bayesian classifier is proposed.The signal is processed by multi-window spectral subtraction,the signal-related features are extracted,and the naive Bayesian classifier is used to distinguish the noise signal segment from the target signal segment.Experimental results show that the detection accuracy of this method is higher under different noise environments and signal-to-noise ratios,and the detection accuracy is significantly improved in non-stationary strong noise environments.Aiming at the problem of type identification of audio signals,an audio type identification method based on deep learning is proposed.In this method,the mel spectrogram of the target audio signal and noise signal is used as the input feature of the deep neural network ECAPATDNN to realize the specific type of target audio signal and noise signal.identify.Experimental results show that this method can accurately realize noise signal type recognition;it can realize high-accuracy audio signal type recognition with or without noise interference.Aiming at the problem of audio signal enhancement,an audio enhancement method based on noise type recognition is proposed.According to the results of noise type identification,this method selects the optimal combination of three parameters in the IMRCA-OMLSA method to realize the optimal estimation of the noise power spectrum,and then uses the optimal improvement target audio estimation is carried out on logarithmic spectrum amplitude estimation.Experimental results show that this method can better improve the quality of noisy audio signals,especially the intelligibility of noisy audio signals.