Abstract: A Method for Implementing Voice-Based Avatar Facial Expression. The method comprises: obtaining, by a processor included in a computing device, chunks based on a voice signal, wherein the chunks comprise an amplitude information of the voice signal for a particular time, inputting input signals corresponding to the chunks into a neural network, computing, by a layer included in the neural network, a value reflecting a relationship between at least two of the input signals based on the input signals, predicting, by an output layer included in the neural network, facial expression coefficients by outputting output signals based on the value and implementing an expression of a predetermined avatar based on the facial expression coefficients. At least two of the output signals are generated by reflecting a relationship between at least two of the input signals.