Non-human voice filtering method in noisy environment based on neural network and MFCC

The invention provides a non-human voice filtering method in a noisy environment based on a neural network and an MFCC. The method comprises the steps: obtaining voice signals which do not contain a human voice and voice signals which contain the human voice, segmenting and supplementing MFCC features, and Delta information, Delta-Delta information and RMSE of the MFCC features, and training the neural network to obtain a trained prediction model; and after a real-time to-be-filtered voice signal is accessed, using the prediction model to score the possibility of human voice existence in each slice according to the MFCC fusion feature of the signal, and finally filtering the non-human voice according to a threshold value. According to the invention, the MFCC-based multi-dimensional features are fused and complemented, and the probability of each voice slice is scored by the neural network so that the problems of low recognition accuracy and incapability of detecting the tail voice of the signal in the noisy e