Voice signal enhancing method and device

A voice signal enhancing method, comprising:

• acquiring a voice signal in the present scene;
• dividing the voice signal into frames according to a preset time interval to generate multiple frame signals;
• feeding the multiple frame signals into a trained neural network according to a preset step size, and performing convolution operations on the multiple frame signals through skip-connected convolutional layers to obtain multiple enhanced frame signals;
• and
• superimposing each enhanced frame signal according to the time domain of each enhanced frame signal to obtain the enhanced voice signal;
• wherein the trained neural network consists of twelve successive convolutional layers, each convolutional layer is followed by a batch normalization layer and finally is activated by a rectified linear unit activation function;
• channels of each convolutional layer are centered on the sixth convolutional layer and symmetrically arranged on both sides of the sixth convolutional layer;
• symmetric convolution pairs of the second convolutional layer and the tenth convolutional layer, symmetric convolution pairs of the third convolutional layer and the ninth convolutional layer, symmetric convolution pairs of the fourth convolutional layer and the eighth convolutional layer, and symmetric convolution pairs of the fifth convolutional layer and the seventh convolutional layer, are skip connected respectively.