Generalized regression neural network sound signal identification method using Mel frequency cepstrum coefficient

The invention relates to a generalized regression neural network sound signal identification method using Mel frequency cepstrum coefficients, which combines MFCC and GRNN, gives full play to the advantages of rich sound features of the MFCC and nonlinear fitting of the GRNN, and effectively identifies seal types. Firstly, MFCC features of sound signals are extracted, FFT and Mel filtering are carried out, an L-order MFCC is solved, cepstrum difference parameters are calculated, a GRNN model is tested, a k-fold cross validation method is used for determining an optimal expansion factor, training data are divided into k folds in the method, the k folds serve as a validation set in sequence for testing, the obtained optimal expansion factor is used for GRNN training, and a GRNN model is obtained. And the test sound data are identified. The reduction of the signal-to-noise ratio has the minimum influence on the GRNN method, when the signal-to-noise ratio is more than 5dB, the GRNN method can realize accurate ide