Multisequence and Multistream Based Derivations of Re-estimation Formulae for Acoustic Model
Aiming at the current situation of lacking literatures presenting systematic description of complete derivation procedures of HMM parameter reestimation formulae for speech recognition, considering discrete and continuous HMM respectively, and taking into account such complex conditions as multiple observation sequences, multiple data streams and mixture Gaussian distributions, we present the complete derivation procedures of HMM re-estimation formulae for speech recognittori which are practically being used widely. We believe that the derivation procedures will certainly help the researchers to gain further insight into the relationship between the speech recognition theory and practice, and will be beneficial for their systematic understanding and mastering-of the stateof-the-art speech recognition technologies.
speech recognition HMM re-estimation formula derivation multisequence multistream
Xirimo Bao Guanglai Gao Xiulian Liu
School of Computer Science Inner Mongolia University Huhhot, China Network Administration Center Ordos Vocational College Ordos, China
国际会议
重庆
英文
258-263
2011-08-20（万方平台首次上网日期，不代表论文的发表时间）