First, this paper introduces the design a corpus based on Uighur news of Xinjiang Broadcast Station. And then, Training the HMM for this corpus. In this step, introduces Word Segmentation, Feature Extraction, Vector Quantization, Codebook, HMM Training etc. Finally, use the corpus and training model implements HMM based Uyghur broadcast news continues speech Sensitive-word spotting and give the conclusion of the test.