论文部分内容阅读
提出一种基于超音段韵律特征和GMM-UBM-MAP的文本无关的说话人确认系统,并与基于MFCC特征参数的说话人确认系统融合,研究提出新的两系统融合策略。在超音段中提取基于基频的韵律特征参数,建立辅助系统。融合时,以基准系统基于MFCC特征参数的说话人确认系统为主系统,基于韵律特征参数的系统为辅助系统,当主系统的得分与阈值接近时,将两系统得分融合再判断。通过NIST2006数据库的实验表明,融合系统相对原系统有16.39%的提升。
This paper proposes a text-independent speaker verification system based on prosody features of supersonic segments and GMM-UBM-MAP, and fuses with the speaker verification system based on MFCC feature parameters to propose a new two-system fusion strategy. Extracting the prosodic feature parameters based on the fundamental frequency in the supersonic section and establishing the auxiliary system. When fusion, the system based on the MFCC feature parameters of the benchmark system is regarded as the main system, and the system based on the prosodic feature parameters is the auxiliary system. When the score of the main system is close to the threshold, the two system scores are fused and judged again. The experiment of NIST2006 database shows that the fusion system has an increase of 16.39% compared with the original system.