Welcome![Sign In][Sign Up]
Location:
Search - MLLR

Search list

[Other5

Description: 语音识别中的说话人自适应研究.nh 1.MAP和MLLR算法比较 文章在讨论由说话人引起的声学差异基础上,研究两种基于模型 的自适应算法:最大似然线性回归(州压LR)和最大后验概率(MAp)。 实验结果表明,不论采用哪种自适应都能使识别率有一定的提升。两 种算法之间的差异性在于MAP具有良好的渐进性,但收敛性较差, 而MLLR在很大程度上改善了收敛特性,但其渐进特性却不如MAP。 文章讨论了在侧汰P自适应中,初始模型参数的先验知识对自适 应效果的影响,以及在MLLR中,回归类对自适应效果的影响。文 章还进一步研究了采用两种算法的累加自适应效果,从结果看MAP 和MLLR结合的方法比单独使用M[AP和MLLR的效果要好。文章 还对包括基于特征层的归一化算法和用于基于声学模型的MLLR算 法等效性进行讨论,并给出了统一的算法框架。-speech paper,help you study
Platform: | Size: 5208064 | Author: 海豚 | Hits:

[Speech/Voice recognition/combineHMM

Description: :为了使应力变异在顽健语音识别系统中能够达到较好的识别效果,研究了基于隐马 尔可夫模型(HMM)的自适应技术,提出了将最大后验概率(MAP)和最大似然回归方法(MLLR)用 于应力变异语音的自适应中。实验结果表明,与基本系统相比,两种方法均有效地提高系统识别 率。以SD为初始模型的最大后验概率方法在150个训练样本时识别效果最好,可以达到90.4% 。-: In order to stress variation in the robustness of speech recognition system can achieve better recognition results, based on Hidden Markov Model (HMM) of adaptive technology, put forward a maximum a posteriori probability (MAP) and Maximum Likelihood regression (MLLR) for the stress of the adaptive variation in voice. The experimental results show that compared with the basic system, both methods are effective to improve the system recognition rate. SD as the initial model to the maximum a posteriori probability method in 150 training samples to identify the best, can reach 90.4 .
Platform: | Size: 234496 | Author: 尹江波 | Hits:

[Othermllr

Description: 语音识别最大似然线性回归(MLLR)算法-Speech recognition maximum likelihood linear regression (MLLR) algorithm
Platform: | Size: 1024 | Author: lixiaofei | Hits:

[Speech/Voice recognition/combineadaptation

Description: 语音识别自适应训练原理的论文,包括MLLR,MAP-adaptation for speech recognition
Platform: | Size: 987136 | Author: 梁伟文 | Hits:

[Speech/Voice recognition/combineSpeech_Adaptin_using_MLLR_and_MAP

Description: 说话人自适应是提高非特定人语音识别系统识别性能的有效手段,本文针对非母语说话人,结合常规的白适应技术MLLR和MAP,探索云南纳西族和傈僳族两种母语说话人的汉语普通话语音识别问题,实验结果显示有显著效果。-Speaker adaptation is a powerful means of improving the performance of speakerindependent speech recognition system.Aimed at Yunnan minority Naxi and Lisu speech,non— native mandarin speech recognition is discussed aplflying general speaker adaptation MLLR and MAP.It is proved to be effective by the experiment results.
Platform: | Size: 363520 | Author: 自然快乐 | Hits:

CodeBus www.codebus.net