9th Annual Conference of the International Speech Communication Association

Brisbane, Australia
September 22-26, 2008

Probabilistic Feature Mapping Based on Trajectory HMMs

Heiga Zen, Yoshihiko Nankaku, Keiichi Tokuda

Nagoya Institute of Technology, Japan

This paper proposes a feature mapping algorithm based on the trajectory GMM or trajectory HMM. Although the GMM or HMM based feature mapping algorithm works effectively, its conversion quality sometimes degrades due to the inappropriate dynamic characteristics caused by the frame-by-frame conversion. While the use of dynamic features can alleviate this problem, it also introduces an inconsistency between training and mapping. The proposed algorithm can solve this inconsistency while keeping the benefits of the use of dynamic features, and offers an entire sequence-level transformation rather than the frame-by-frame conversion. Experimental results in voice conversion show that the proposed algorithm outperforms the conventional one both in objective and subjective tests.

Full Paper

Bibliographic reference.  Zen, Heiga / Nankaku, Yoshihiko / Tokuda, Keiichi (2008): "Probabilistic feature mapping based on trajectory HMMs", In INTERSPEECH-2008, 1068-1071.