Voice recognition method and device

Disclosed are a voice recognition method and device. The method comprises: receiving a voice signal (S101); decoding the voice signal according to an acoustic model, a linguistic model and a decoding network established in advance, and dynamically adding a blank unit in a decoding process to obtain an optimum decoding path with the added blank unit, wherein the acoustic model is obtained based on connecting time-sequence classification training, the acoustic model comprises a basic pronunciation unit and the blank unit, and the decoding network consists of a plurality of decoding paths composed of the basic pronunciation unit (S102); and outputting the optimum decoding path as a recognition result of the voice signal (S103). The voice recognition method can improve the accuracy of voice recognition, and improve the decoding speed in a recognition process.