The embodiment of the invention provides a voice recognition method and device, equipment and a medium. The method comprises the steps: converting acquired audio data into a corresponding spectrogram;judging whether the frame number of the spectrogram is a preset frame number or not; if the frame number of the spectrogram is not the preset frame number, carrying out zero filling on the spectrogram, so as to making the frame number of a spectrogram to be identified obtained after zero filling is the preset frame number; and inputting the spectrogram to be identified into a multi-task convolutional neural network acoustic model. The spectrogram is directly input into an acoustic model, and then the text of the audio data is recognized. Compared with the prior art for calculating the information loss on the frequency domain caused by MFCC characteristics, the method and device have the advantages that the loss of the input characteristics is reduced, the identification degree of the audio data is increased, and the extraction of the characteristic information by the acoustic model is facilitated.

Wu Weidong

Papers

Speech recognition method and device, equipment and medium