I would imagine this data set of four to five thousand hours of traditional Chinese Mandarin is not a public domain, licensable data set.