ITRW on Non-Linear Speech Processing (NOLISP 07) Paris, France May 22-25, 2007

Tone Recognition in Mandarin Spontaneous Speech

Zhaojie Liu (1,2), Pengyuan Zhang (1), Jian Shao (1), Qingwei Zhao (1), Yonghong Yan (1) Ji Feng (2)

(1) ThinkIT Speech Lab, Institute of Acoustics, Chinese Academy of Sciences; (2) Institute of Physics, Chinese Academy of Sciences, Beijing, China

This paper reports our study on tone recognition in Mandarin spontaneous speech, which is characterized by complicated tone behaviors. Real-Context is proposed as a new concept used in the tone modeling. First, the(1)ata, which may bring negative influences to the tone model, are removed from the training data by an iterative method. Then we cluster the reduced training data into a few subsets to generate a more refined tone model. Gaussian Mixture Model (GMM) is used for the tone modeling. All experiments are based on the spontaneous speech database, Train04. Experimental results demonstrate the effectiveness of the methods.

Full Paper

Bibliographic reference.  Liu, Zhaojie / Zhang, Pengyuan / Shao, Jian / Zhao, Qingwei / Feng, Yonghong Yan (1) Ji (2007): "Tone recognition in Mandarin spontaneous speech", In NOLISP-2007, 100-103.