4th International Conference on Spoken Language Processing

Philadelphia, PA, USA
October 3-6, 1996

Adaptive Recognition Method Based on Posterior Use of Distribution Pattern of Output Probabilities

Jin-Song Zhang (1), Beiqian Dai (1), Changfu Wang (1), Hingkeung Kwan (2), Keikichi Hirose (2)

(1) Dept. of Elec. Engr., USTC, Hefei, Anhui, PRC
(2) Dept. of Elec. Infm., Univ. of Tokyo, Hongo, Bunkyo-ku, Tokyo, Japan

We propose a new adaptation scheme for speaker independent recognition. The basic idea lies in the change of the likelihood from the ordinary HMM scores to the combined observational scores. The new likelihood is computed based on a combination of HMM scores which we called a Distribution Pattern of Output Probabilities (POPD). The system needs to calculate only the POPD for each new speaker. Re-estimation of acoustic model parameters is unnecessary. Preliminary experiments on Chinese icolated-syllable recognition indicate the methodís effectiveness.

Full Paper

Bibliographic reference.  Zhang, Jin-Song / Dai, Beiqian / Wang, Changfu / Kwan, Hingkeung / Hirose, Keikichi (1996): "Adaptive recognition method based on posterior use of distribution pattern of output probabilities", In ICSLP-1996, 1129-1132.