Duration refinement by jointly optimizing state and longer unit likelihood

Boyang Gao, Yao Qian, Zhizheng Wu, Frank K. Soong