发明名称 FULL-SEQUENCE TRAINING OF DEEP STRUCTURES FOR SPEECH RECOGNITION
摘要 <p>A method is disclosed herein that include an act of causing a processor to access a deep-structured model retained in a computer-readable medium, wherein the deep-structured model comprises a plurality of layers with weights assigned thereto, transition probabilities between states, and language model scores. The method can further include the act of jointly substantially optimizing the weights, the transition probabilities, and the language model scores of the deep-structured model using the optimization criterion based on a sequence rather than a set of unrelated frames.</p>
申请公布号 EP2619756(A2) 申请公布日期 2013.07.31
申请号 EP20110827195 申请日期 2011.09.07
申请人 MICROSOFT CORPORATION 发明人 YU, DONG;DENG, LI;MOHAMED, ABDEL-RAHMAN SAMIR ABDEL-RAHMAN
分类号 G10L15/14;G10L15/16 主分类号 G10L15/14
代理机构 代理人
主权项
地址