摘要: 定义了10种基本的嘴形。以Mel频率倒谱系数(MFCC)作为语音特征,通过SVM分类器进行元音a,i,u的识别,根据其对应量化后的语音能量,映射到嘴形序列,进行中值滤波和排除“奇异点”。该算法在基于语音驱动人脸动画系统中的应用取得了良好的效果。
关键词:
语音,
嘴形,
Mel频率倒谱系数,
能量
Abstract: This paper defines 10 kinds of mouth shapes. The MFCC feature of speech is extracted to recognize vowel (a, i, u) through SVM classifier. According to corresponding quantified speech energy, the speech is mapped to mouth shapes sequences, and median algorithm filter and “odd point” eliminating are undertaken. The algorithm has been applied to the system of face animation driven by speech and achieves good results.
Key words:
speech,
mouth shape,
MFCC,
energy
中图分类号:
林 鑫;陈 桦;王开志;王继成. 语音驱动唇形自动合成算法[J]. 计算机工程, 2007, 33(17): 237-238,.
LIN Xin; CHEN Hua; WANG Kai-zhi; WANG Ji-cheng. Automatic Lip Synchronization Algorithm Driven by Speech[J]. Computer Engineering, 2007, 33(17): 237-238,.