The invention discloses a voice identification method. The voice identification method comprises steps that in the continuous voice signal acquisition process, a custom corpus corresponding to a present account is acquired; the custom corpus is analyzed and processed, and at least one corresponding custom decoding model is constructed; the at least one corresponding custom decoding model is loadedto a pre-stored universal decoding model, and a new decoding model is generated; the new decoding model is employed to decode a voice signal to acquire the voice identification result. The inventionfurther provides a voice identification device and a voice identification engine.