CSLT released a free Chinese speech database, THCHS-30. This database was recorded by Dr. Dong Wang in 2000-2001 when he was a master student at Tsinghua university, supervised by Prof. Xiaoyan Zhu. The database involves more than 30 hours of speech signals, and additional resources such as lexica, LMs, noise data are provided. This forms a full set of resources that can be used to build a completed Chinese speech recognition system.
Check the database homepage for details.