The speech recognition system is not real-time, a speak identification method based on the cloud server is proposed to solve this problem. Firstly, the MFCC frequency cepstrum coefficient and the first order differential coefficient are extracted from the speech feature vector sequence to form 32 dimensional. And then the 32 dimensional speech feature vector is sent to the cloud server, the training speaker model and identification are done in the cloud server. Finally, the identification result is sent to the client.