Speech-Music Classification Model Based on Improved Neural Network and Beat Spectrum

A speech-music classification method according to a developed neural system and beat spectrum is proposed achieve accurate of through pre-emphasis, endpoint detection, framing, windowing other steps preprocess collect vocal music signals. After fast Fourier transforms triangle filter processing, the Mel frequency cepstrum coefficient (MFCC) obtained, discrete cosine transform performed obtain signal MFCC characteristic parameters. calculating similarity feature parameters similarity, matrix based on which obtained. The residual structure optimized by adding Swish max-out activation functions, respectively, between convolutional network layers build convolution deepen number layers. connected time series (CTC) used as objective loss function. It applied softmax layer deep optimization for model. pitch input information model realize classification. experiment proves that accuracy design higher than 99%; when iteration reaches 1200, training approaches 0; signal-to-noise ratio 180dB, sensitivity specificity are 99.98% 99.96%, respectively; voice 99%, running 0.48 seconds. has been proven high accuracy, low loss, good special effects, can effectively speech-music.