Design of Political Online Teaching Based on Artificial Speech Recognition and Deep Learning

Table 1

CTC-CNN baseline acoustic model parameters.

Network layer	Parameter
InputLyer	300 dimensional harmony
conv2d_Lyer1	33 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
conv2d_Lyer2	33 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
max-poolng2d	Maximum pooling 3 × 3
conv2d_Lyer3	65 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
conv2d_Lyer4	65 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
max-poolng2d	Maximum pooling 3 × 3
conv2d_Lyer5	129 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
conv2d_Lyer6	129 convolution channels, convolution kernel 4 × 4, step size 3 × 3, activation function
max-poolng2d	Maximum pooling 3 × 3
Reshap	Feature map transformation output 300 × 3300
FC_Lyer1	Number of neurons 129, activation function
FC_Lyer2	The number of neurons is 1425, the activation function
Softmx	Activation output matrix, the dimension is 300 × 1425
CTC	Probability matrix, the length of the sonogram, the length of the label sequence