Speech recognition method and device

The invention discloses a speech recognition method and device. The speech recognition method comprises steps: a first neural network is used to extract vectors containing recognition features of a speaker from to-be-recognized speech data; according to the vectors containing the recognition features of the speaker, bias compensation is carried out on a second neural network; and through an acoustic model based on the second neural network, the to-be-recognized speech data are subjected to speech recognition. In a condition of not introducing an excessive amount of calculation, the performanceof adaptive speech recognition of an online speaker can be effectively enhanced.